Guidebook on Pedestrian and Bicycle Volume Data Collection (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

75 c h a p t e r 5 This chapter summarizes existing and emerging sensor technologies available for non- motorized counting. Many of the technologies were developed for motor vehicle counting but have been adapted for non-motorized travel. Other technologies are more specific to non- motorized counting. Each counting technology or method (in the case of manual counting) is presented in its own subsection, along with the following information: • Description of how the counting technology or method detects pedestrians or bicyclists. • Typical applications for the technology. • General installation considerations for the technology. Manufacturer’s installation recom- mendations take precedence over these general considerations. • Relative level of effort and cost, drawing from the literature, vendor-provided information, and the research team’s experience. Specific product costs are subject to change and the cost of additional services (e.g., pavement cutting) vary greatly by region, size of the order (e.g., consideration of economies of scale), and whether or not an organization can perform the work in house or has to contract it out. Sensor Technology Toolbox Chapter 5 Topics Chapter 5 summarizes typical applications, installation considerations, relative level of effort and cost, strengths and limitations, accuracy, and usage of 14 existing and emerging counting technologies and methods: • Manual in-field counting • Manual counts from video • Automated counts from video • Pneumatic tubes • Inductive loop detectors • Passive infrared • Active infrared • Piezoelectric strips • Radio beams • Thermal • Laser scanners • Pressure and acoustic pads • Magnetometers • Fiberoptic pressure sensors

76 Guidebook on pedestrian and Bicycle Volume Data collection • Strengths and limitations of the technology or method, drawing from the literature and the research team’s experience. • Accuracy, drawing from the NCHRP Project 07-19 testing when possible, and supplementing from the available literature. • Description of current usage, drawing from the NCHRP 07-19 practitioner surveys and interviews. 5.1 Manual In-Field Counts 5.1.1 Description Human data collectors can record pedestrian and bicycle volumes using paper sheets, traf- fic count boards, “clicker” counters, or smartphone apps. Counts are usually recorded for 1 to 4 hours in discrete time intervals, generally 15 minutes, although counts can be collected in shorter intervals if desired. Some count boards can timestamp all data points. Data collector training, motivation, and management are important for obtaining accurate manual counts. 5.1.2 Typical Applications Manual counts can capture both pedestrian and bicycle volumes and can capture other infor- mation as well (e.g., gender, helmet use, or risky behaviors). They can capture both screenline volumes and intersection turning movement and crossing volumes. Manual counts are most appropriate for collecting data over a relatively short time on any facility type. They are one of the few approaches that can capture turning movements or addi- tional information about users. Because of low additional costs per site, manual counts can be used to inexpensively increase the number of sites observed. They are also a good starting point for new count programs, both because of low start-up costs (the only significant cost is labor) and to help in prioritizing sites for installing automated counting equipment. Finally, manual counts are necessary for validating automated counting equipment. 5.1.3 Installation Considerations Before collecting counts, the site should be assessed to determine the specific location(s) at which the manual counter(s) should be positioned to most easily view users. Based on the antici- pated user volumes and the kinds of information that will be collected, more than one person may be needed. A general rule of thumb is that one person should be expected to capture no more than 200 data points per hour (including the actual volumes as well as any additional attributes). The National Bicycle and Pedestrian Documentation Project (NBPD, Alta Planning + Design 2012) recommends that counters working longer than 2 hours be relieved for restroom breaks at least every 2 hours and provided 30-minute lunch periods. Counters should be trained on how to classify users and collect data. Appendix B provides example data collector instruc- tions for collecting manual counts at intersections. In addition, the NBPD provides training materials and sample count and survey forms at bikepeddocumentation.org. 5.1.4 Level of Effort and Cost The costs of conducting manual counts are largely determined by the number of hours of data that are collected. Data collectors (either staff or volunteers) must be trained, but mul- tiple people can be trained simultaneously. Sections 3.2.2 and 3.3.4 discuss the potential benefits and disadvantages of using volunteers to conduct counts and stress the importance of training Source: Robert Schneider, UC Berkeley SafeTREC. Counters using data collection sheets.

Sensor technology toolbox 77 volunteers. Hand-written data sheets, if used, must be gathered, compiled, and reduced into a spreadsheet, which can be time-intensive. (Count boards, smartphone apps, and similar tools can export the data into a spreadsheet-compatible format.) Labor costs typically include training time, travel time to and from the count site, and preparation time (counters should arrive on site early to orient themselves), in addition to the actual time required to perform the counts. Two person-hours are generally required per 1 hour of counts performed (including preparation time and training), plus additional time if the data must be manually entered into a spreadsheet or database. Additional person-hours are required at high-volume locations and when additional user information is being collected. The cost of labor drives the overall cost per count. This is in contrast to most automated count technologies, where the equipment capital cost drives the overall cost. 5.1.5 Strengths and Limitations Strengths • Flexibility to gather additional data about traveler (i.e., directional information, gender, and behaviors). • Applicable to all site types and users. • No installation costs or impacts. • Extremely mobile. Limitations • Short-term counts only. • More personnel needed at higher volume locations or to collect additional informa- tion (i.e., higher costs). • Subject to data collector fatigue and (with volunteers) possible count biases. Manual Count Summary Maximum user volume: Up to 600 persons per hour per counter Detection zone width: >75 feet Typical count duration: 4 hours or less Typical equipment cost (2013): <$1,000 Relative preparation cost: Low Typical installation time: Negligible, beyond arriving on site early to orient one’s self Typical data collector training time: >1 hour Relative hourly cost: Very high, can exceed $50/hour for training, management, and on-site labor costs Mobility: Excellent, no assembly required for human counters 5.1.6 Accuracy Accuracy depends on data collector behavior. It improves with training and experience and decreases with count duration and quantity of additional information collected. One study of manual intersection counts at busy intersections in San Francisco found undercount rates rang- ing between 8% and 25% for 15-minute intervals, compared to manual counts conducted on video footage (Diogenes et al. 2007). Accuracy was worse at the beginning and end of the data collection period, which is likely attributable to a familiarization interval (which can be miti- gated by more on-site time prior to beginning counts) and fatigue.

78 Guidebook on pedestrian and Bicycle Volume Data collection 5.1.7 Usage The NCHRP Project 07-19 practitioner survey found that manual counts are the most preva- lent approach in the United States for collecting non-motorized volumes. Of respondents who performed counts, 93% included manual counts as part of their pedestrian data collection pro- gram and 87% included them as part of their bicycle data collection program. Most respon- dents used manual counts taken annually at strategically chosen and distributed locations, using agency staff, volunteers, contractors, or some combination of these. 5.2 Manual Counts from Video 5.2.1 Description Manual counts can be made from video footage collected with a temporarily or permanently installed camera. Videos are reviewed manually on a monitor after they are collected, with the data collector counting using a paper sheet, a handheld counter, or a computer. Specialized key- boards are available commercially that can be plugged directly into a computer. Manual counts from video were used to develop the ground truth counts in the NCHRP Proj- ect 07-19 research. The process used is described in the project’s final report (Ryus et al. 2014). 5.2.2 Typical Applications Manual counts on video can be used to capture pedestrian and bicycle volumes, including crossing counts and turning movement counts at intersections. They might be useful for observ- ing individual characteristics (e.g., gender and helmet usage), although these details can be hard to discern in lower resolution video images or in images taken from a distance. Specific behaviors that would be missed by most automated technologies can be observed (e.g., wrong-way riding, traffic control device compliance, and sidewalk riding). Source: Frank Proulx, UC Berkeley SafeTREC. Technician installing video camera. Manual Count from Video Summary Maximum user volume: More than 600 persons per hour per counter, but higher volumes require more data-reduction time (e.g., slower playback speed and more frequent need to rewind) Detection zone width: >75 feet Typical count duration: 4 hours or less, but longer counts can be performed by spreading the work over multiple data reduction sessions Typical equipment cost (2013): <$1,000 Relative preparation cost: Low Typical installation time: <30 minutes Typical data collector training time: >1 hour Relative hourly cost: Very high, can exceed $50/hour for training, management, and labor costs Mobility: Very good, only a camera needs to be installed

Sensor technology toolbox 79 5.2.3 Installation Considerations Video cameras should be installed high enough above the street (e.g., 10+ feet) to deter theft. They should also be mounted in inconspicuous cases that are resistant to theft or in vandalism-resistant cases. When mounting the camera, make sure that the desired detection zone is being clearly recorded and is not likely to be obscured (e.g., by tree branches moving in the wind or by stopped trucks or buses). Site visits are typically required every 2 to 3 days when video is being collected to ensure the camera is working properly, swap memory cards, and replace camera batteries as needed. 5.2.4 Level of Effort and Cost Consumer-grade camcorders can be mounted in secure boxes for a low-cost solution. Higher quality cameras can be used to collect more detailed video or to remotely transfer video. A monitor or computer capable of video playback is also needed. Computer software to assist with recording counts may also be helpful. Equipment cost is typically lower for this method than for automated counters; however, labor costs are also required for video reduction and camera set-up, maintenance, and take-down, and to periodically conduct quality-assurance checks of the counts. As with manual in-field counts, the cost of reducing video data is largely determined by the number of hours of data collected. Appendix C provides the protocol used by NCHRP Project 07-19 to develop manual counts from video. 5.2.5 Strengths and Limitations Strengths Limitations • Flexibility to gather additional data about users (i.e., directional information, gender, and behaviors). • Applicable to all site types and users. • Flexibility to slow down or speed up video data during reduction based on volume of users. • Data collectors do not have to spend hours in the field—a constraint during poor weather conditions or nighttime data collection. • Video can be reviewed at times other than when data are collected to accommodate busy schedules. • A single data collector can reduce data for the same time period at multiple sites after video cameras are set up. • Short-term counts only, because of the labor costs involved with reducing data. • Frequent field visits are necessary to set up cameras, replace batteries and memory cards, and take down equipment. • Video cameras are susceptible to theft unless well-obscured and placed out of convenient reach. • Problems can arise with video footage (e.g., corrupt files or poor vantage points) requiring the video to be retaken. • Requires a fixed pole at the location or a portable pole for mounting the camera. 5.2.6 Accuracy Video-based manual counts are presumed to be the most accurate way of collecting count data, given the ability to re-watch video data and to slow down the playback speed as needed. How- ever, objects that block the camera’s field of view (e.g., large vehicles such as buses and trucks) can result in missed detections, particularly if they stop for a period of time. No controlled tests (specifying the number of people using a facility) appear in the literature, and such tests would be difficult to arrange for large volumes. This method was used to develop the ground truth validation counts for NCHRP Project 07-19.

80 Guidebook on pedestrian and Bicycle Volume Data collection 5.2.7 Usage The NCHRP Project 07-19 practitioner survey found that 44% of respondents who performed pedestrian counts used manual counts from video data as part of their pedestrian data collec- tion program. (The question was not asked in conjunction with bicycle data collection; however, given the similar pedestrian and bicycle counting responses for other counting methods and technologies, the proportion would be expected to be similar.) 5.3 Automated Counts from Video 5.3.1 Description Pedestrians or bicyclists are counted from video images by using computer algorithms to identify when changes in the background image are pedestrians or bicyclists passing through the detection area. 5.3.2 Typical Applications Automated counts from video can capture pedestrian or bicycle volumes. They can capture both screenline volumes and intersection turning movement volumes, although it may take multiple cameras to collect data from an entire intersection. Automated video collection is typically used for up to 1 week at a time, because of data storage limitations for video images. 5.3.3 Installation Considerations Mount the camera high enough to capture the desired area, using existing infrastructure if possible, but trying to avoid sources of vibration (e.g., traffic signal mast arms). Typically, a mounting height of approximately 25 feet is required. The camera will need to be placed within a certain distance of the roadway (e.g., 12 feet) and can be expected to capture a detection zone Source: Ling et al. (2010). Detection of pedestrians using automated video counting. Automated Video Count Summary Maximum user volume: >600 persons per hour Detection zone width: >75 feet Typical count duration: ≤48 hours Typical equipment cost (2013): $1,000–3,000 Relative preparation cost: Medium Typical installation time: <30 minutes Typical data collector training time: <30 minutes Relative hourly cost: High, at the time of writing, this technology was only available as a vendor-supplied service Mobility: Very good, only a camera needs to be installed

Sensor technology toolbox 81 of a certain size (e.g., out to 150 feet). Therefore, multiple cameras may be required to capture an entire intersection. Care should be taken to avoid environmental conditions that could affect results, such as glare from nearby streetlights or the sun. 5.3.4 Level of Effort and Cost The level of effort is medium relative to other technologies and requires setting up and taking down the video camera and sending the video to the vendor for processing. The one commercial system in the North American market at the time of writing operated as a service, meaning that videos were sent to the vendor to be processed by the vendor’s system. According to the vendor, its staff reviewed the data produced by the automated system. The cost per count is medium to high, relative to other technologies. 5.3.5 Strengths and Limitations Strengths Limitations • Minimal human time required to collect counts. • Can provide intersection turning movement and crosswalk counts. • Short-term counts only. • Not currently possible to process video in-house (requires a vendor to do the processing). • Portable and straightforward to install where camera mounting locations are available. • Video can be used for additional purposes (e.g., facility evaluation and user behavior studies). 5.3.6 Accuracy The accuracy of the automated video counting service available in the marketplace has not been rigorously tested. However, one could perform one’s own checks by making manual counts from a sample of videos sent to the vendor for processing and comparing one’s results to the vendor’s results. 5.3.7 Usage Of those project practitioner survey respondents who performed counts, 18% used auto- mated video as part of their pedestrian counting program and 17% used it as part of their bicycle counting program. Much of the work to develop improved systems for collecting automated counts from video is still in the research stage and not yet commercially available. 5.4 Pneumatic Tubes 5.4.1 Description This technology is applied by stretching one or more rubber tubes across the roadway or path- way. Although general purpose counters (GPCs) typically used for motorized vehicle detection and classification can be used to count bicyclists, specialized bike-specific counters (BSCs) are also available, which only count bicyclists. When a bicycle or other vehicle passes over a GPC tube, a pulse of air passes through the tube to a detector, which then registers a count. Multiple tubes can be used to determine speed and directionality. In some cases, the number of axle hits is divided by two to deduce the number of bicyclists. In other cases, more complicated classification algorithms Source: Karla Kingsley, Kittelson & Associates, Inc. Bicyclist riding over pneumatic tubes.

82 Guidebook on pedestrian and Bicycle Volume Data collection can be applied to the pattern of axle hits to determine the vehicle speed and classification. This process can either occur in real time or on a personal computer after the data have been recorded. BSCs designed for bicyclists can be used on bike-specific facilities or in mixed traffic. When used in mixed traffic, the difference in the air pulse created by heavier motorized vehicles can be detected and disregarded, so that only bicyclists are counted. Tubes designed for bicyclists are generally smaller, to minimize the bump as cyclists ride over them, and are more sensitive, to better detect bicyclists. 5.4.2 Typical Application Pneumatic tubes can be very effective when bicycle data need to be collected for several days up to several weeks. Tubes are most appropriate for paved surfaces with minimal pedes- trian use and temperatures above freezing, because tubes may not maintain their properties in cold weather and can deteriorate. Tubes can be used on bike-specific facilities or in mixed traffic. When used on street, tubes can be placed just across the bike facility or across the entire roadway. If only bicycle counts are needed, placing BSCs across just the bike facility is preferred, given bicyclist comfort considerations and reduced wear and tear on the tubes. If both motorized traffic volumes and bicyclist volumes are required, than placing GPCs across the entire roadway is necessary. 5.4.3 Installation Considerations Tubes are installed across the paved surface in a location where bicyclists are unlikely to stop. The tube should adequately cover the bicycle travel path, while minimizing exposure to motor vehicles. Locations and times when street sweeping or snowplowing occurs should be avoided, because those vehicles can dislodge or destroy tubes. Tubes can be affixed to the roadway using either mastic tape or nails and brackets. Some jurisdictions may oppose having nails in their roads, because of concerns about pavement damage. Care should be taken to consider bicyclist safety and to minimize the risk of a nail or metal fixture puncturing a bike tire, by placing metal objects outside the bicycle facility or by using tape to secure the tubes. Pneumatic Tube Summary Maximum user volume: Provides consistent results up to 200 users per hour; counts can be corrected at higher volumes Detection zone width: <20 feet Typical count duration: Non-permanent short- and longer-term counts Typical equipment cost (2013): $1,000–3,000 Relative preparation cost: Medium (potential need for permits) Typical installation time: <30 minutes Typical data collector training time: <30 minutes Relative hourly cost: Medium, equipment costs are spread over more data-collection hours than for manual counts Mobility: Very good, equipment can be removed and taken to a new site

Sensor technology toolbox 83 Follow the manufacturer’s instructions on spacing and settings. If GPCs are being used, the settings need to be modified to apply a classification scheme to sort vehicles and isolate bicycle data. The manufacturer may suggest appropriate settings; research is being conducted to develop classification schemes to improve pneumatic tube accuracy. 5.4.4 Level of Effort and Cost The level of effort is low, relative to other technologies, and requires setting up and taking down the tubes. Most jurisdictions are likely already familiar with the set-up process because of their experience with pneumatic tubes for counting motorized vehicles. The cost is typically low to set up the equipment and process the counts. 5.4.5 Strengths and Limitations Strengths • Portable. • Easy to set up. • Can capture speed and directionality of bicyclists when two tubes are used. • Most jurisdictions are familiar with the technology, because they already use it for counting automobiles. Limitations • Susceptible to theft, vandalism, dislodge- ment, and wear and tear, requiring routine maintenance. • May require permission from local juris- diction for installation, which sometimes requires not using nails. • May not maintain properties in very cold conditions and can deteriorate under high- traffic conditions. One vendor of BSCs claims to have observed their tubes to last for approximately 300,000 vehicle hits. • Not usable during times when street sweeping or snowplowing occurs, because the tubes can be dislodged or destroyed. 5.4.6 Accuracy NCHRP Project 07-19 tested two models of BSC tubes, primarily on multiple-use paths and bicycle lanes. The testing found that the tubes generally undercounted, but that over- counting sometimes occurred at higher volumes. One tested model performed substantially better than the other model, with a correction factor of 1.127 versus 1.520 (a factor of 1.00 indicates that no correction is needed). The average percentage deviation (APD), represent- ing the overall divergence from perfect accuracy across all data collected, was -17.9%. The average of the absolute percent difference (AAPD), representing the counter’s consistency, was 18.5%. Figure 5-1 shows the accuracy and precision of the pneumatic tubes tested by NCHRP Project 07-19. Research from a field test of bicycle counts using two pneumatic tube models in New Zealand found a range of accuracies for off-road locations from -15% to 0%, with the tubes typically undercounting bicyclists (ViaStrada 2009). Hjelkrem and Giæver (2009) tested two models of pneumatic tubes in mixed traffic and found bicycle count accuracy rates of -27.5% and -1.9%. Hyde-Wright, Graham, and Nordback (2014) compared the accuracy of BSCs to GPCs. They found that the BSC proved very reliable and accurate when counting bicyclists striking the pneu- matic tubes up to 27 feet away from the counter (with an average accuracy between 94 and 95%). However, accuracy decreased for bicyclists at greater distances from the counter, with an accuracy of 57% for bicyclists riding 33 feet from the counter. The accuracy and reliability of GPCs proved variable based on the attachment method and classification scheme used. A custom classification

84 Guidebook on pedestrian and Bicycle Volume Data collection scheme was developed through the study to account for the systematic undercounting observed with the GPCs. Using this classification scheme, the GPCs proved accurate when bicyclists rode close to the counter, with an average accuracy of 95%. Accuracy declined significantly for bicy- clists riding farther from the counter. 5.4.7 Usage Of this project’s practitioner survey respondents who performed counts, 27% used pneumatic tubes as part of their bicycle counting program. The survey did not distinguish between GPCs and BSCs. 5.5 Inductive Loop Detectors 5.5.1 Description Wires are installed under the surface of the pavement (embedded) or on top of the pavement (temporary). Small electrical currents running through the wires that form the loops generate a magnetic field; the sensor detects changes in this magnetic field that occur when metal parts of a bicycle (e.g., frame, spokes, and pedals) pass over the loops. 5.5.2 Typical Application Loop detectors are generally intended for permanent count locations (embedded loops), but can be used for shorter duration counts with temporary surface loops designed for bicycle counting (as shown in the picture to the right). Loop detectors are used to collect screenline counts and are typically used on paved facilities, although at least one vendor makes a product using pre-formed loops that can be buried in soil. Although inductive loop detectors can be used on exclusive bicycle facilities, on mixed-use paths, and in mixed traffic, they have shown to be more accurate in situ- ations where bicycles are separated from motor vehicle traffic (Nordback et al. 2011). Inductive loops used to detect bicycles at traffic signals may also be a potential source of counts, but not all traffic signal controllers can process and store bicycle count data (Kothuri et al. 2012a). Source: NCHRP 07-19 testing. Figure 5-1. Accuracy and precision of tested pneumatic tubes. Technicians installing temporary inductive loop detectors. Source: Katie Mencarini, Toole Design Group.

Sensor technology toolbox 85 5.5.3 Installation Considerations Select a mid-segment, channelized location where bicyclists are unlikely to stop and will be more likely to ride single file. Locations where loops can cover all (or nearly all) of the bicycle facility are preferred, as are locations where bicyclists cannot easily bypass the detectors. Embedded loop detectors require pavement sawcutting to install the loops. Depending on the situation, considerable lead time may be required to obtain necessary permits, hire a contractor, and schedule the installation. (This process may be simpler when the agency installing the coun- ter also owns the roadway or facility, and when the agency has in-house sawcutting expertise.) The data logger is typically stored in a utility box adjacent to the facility, which may require some excavation. Temporary loops are adhered to the pavement surface, typically using adhesive tape. It can be difficult to remove the tape at the end of the counting period. Temporary loops are meant to be discarded at the end of service, but the data logger and other hardware can be reused with new sets of loops at other sites. With either type of loop, consideration will need to be given to managing bicycle (and perhaps other) traffic while the installation is occurring. 5.5.4 Level of Effort and Cost Relative to other counting technologies, the cost and level of effort to install embedded loops is high (due to the need for sawcutting and traffic control). The relative cost and level of effort to install temporary loops is medium, given the need for traffic control and ensuring that the wiring is set up correctly. Installation costs for embedded loops can vary greatly, depending on whether the organization installing the counter has the in-house expertise to install loops (e.g., for traffic signals) or needs to hire a contractor. Inductive Loops Summary Maximum user volume: Provides consistent results up to 600 users per hour; counts can be corrected at higher volumes. Detection zone width: <20 feet Typical count duration: Embedded loops are designed for permanent installa- tions, temporary loops can be used for shorter-term counts (<6 months) Typical equipment cost (2013): $1,000–3,000 Relative preparation cost: Moderate to high (potential need for permits and sawcutting/excavation, traffic control) Typical installation time: >half day (embedded loops), several hours (temporary loops) Typical data collector training time: <30 minutes Relative hourly cost: Low, equipment costs are spread over a large number of data-collection hours Mobility: Poor, embedded loops are designed for permanent installations; the data logger used for temporary loops can be moved

86 Guidebook on pedestrian and Bicycle Volume Data collection 5.5.5 Strengths and Limitations Strengths • Most jurisdictions are familiar with embedded loop technology, because it is also used to detect vehicles at traffic signals. • Can be used for on-street bicycle facilities. • Can be battery powered. • Long-lasting equipment. Limitations • Embedded loops require pavement saw cuts and a minimum pavement thickness. • Electromagnetic interference can cause errors. • May not detect side-by-side bicyclists. • May experience inaccuracies with non- standard bicycles (e.g., bicycles with trail- ers or cargo boxes, tandem bicycles). • If it is not possible to cover the entire facil- ity width with the loops, bypass errors will occur when bicyclists ride outside the area covered by the loops. 5.5.6 Accuracy NCHRP Project 07-19 tested one model of embedded loops and one model of temporary loops, primarily on off-street facilities. The two types of loops had similar accuracy rates when used off-street. Overall, the APD, representing the overall divergence from perfect accuracy across all data collected, was -0.6%. The AAPD, representing the counter’s consistency, was 8.9%. Both of these values exclude bypass errors. Figure 5-2 shows the accuracy and precision of the inductive loops tested by NCHRP Project 07-19, based both on (1) only the bicycles passing through the detection zone and (2) including bypass errors, where bicyclists were able to ride around the detection zone. The degree of miscounting because of bypass errors depends on the characteristics of the count site and the degree to which the loops cover the bicycle facility width. Consequently, site-specific correction values should be developed to account for bypass errors. Tests of embedded loop detectors in Colorado showed an accuracy of -4% at off-road loca- tions, -3% accuracy on separated paths, and +4% accuracy on shared roadways (Nordback et al. 2011; Nordback and Janson, 2010). Testing in New Zealand showed ranges of accuracy from Source: NCHRP Project 07-19 testing. (a) Bicyclists Passing Through Detection Zone (b) All Bicyclists Using the Facility Figure 5-2. Comparison of accuracy and precision of inductive loops.

Sensor technology toolbox 87 -10% to +4% for on-road count sites and -10% to +25% for off-road sites (ViaStrada, 2009). At the time of writing, research was ongoing related to the accuracy of counts from traffic signal bicycle loop detectors (Nordback, Johnson, and Koonce 2014). 5.5.7 Usage Embedded loop detectors have been more widely tested and used than temporary loops. Both types of loops are commercially available. Transportation agencies are familiar with embedded loop detectors given their use in vehicle and bicycle detection at signalized intersections. Of those practitioner survey respondents who performed counts, 23% used inductive loops as part of their bicycle counting program. 5.6 Passive Infrared 5.6.1 Description Passive infrared (IR) devices detect pedestrians and cyclists by comparing the temperature of the background to the infrared radiation (heat) patterns emitted by persons passing in front of the sensor. The passive infrared sensor is located on one side of the facility being counted. These devices are also known as “pyroelectric” counters, which refers to how the heat received by the sensor changes the sensor’s electrical properties. 5.6.2 Typical Application Passive IR devices are appropriate for collecting counts for several weeks or as permanent installations. They cannot differentiate between pedestrians and bicyclists, so are best for facilities with one user type or in conjunction with a bicycle-only counting technology to differentiate users. They collect screenline counts and can be used on multi-use paths or sidewalks. Integrated units that combine passive IR (to count all users) with either inductive loops or piezoelectric strips (to count bicyclists) are commercially available. When using a combination unit, the pedestrian count is obtained by subtracting the bicycle count from the total user count. 5.6.3 Installation Considerations The placement of passive IR counters is critical to obtaining good results. IR counters are typically positioned on one side of the count corridor inside a post or placed on existing infra- structure. Passive IR sensors should be placed at the vendor-specified height (typically 2 to 3 feet) and work best when installed pointing toward a fixed object (e.g., a wall). Avoid locations where there is a likelihood that pedestrians will linger or congregate (e.g., doorways, bus stops, or street corners). Care should also be taken to avoid problems with reflection from heavy foli- age, water, windows, or background traffic. Jones et al. (2010) found that the passive IR sensor model they tested produced the best results when positioned at a 45-degree angle to the pathway, to minimize occlusion. 5.6.4 Level of Effort and Cost The level of effort to install a passive infrared detector is low, requiring mounting a single device on one side of the facility. The cost per device is medium, relative to other technologies. Source: Ciara Schlichting, Toole Design Group. Technicians setting up a passive infrared counter.

88 Guidebook on pedestrian and Bicycle Volume Data collection 5.6.5 Strengths and Limitations Strengths • Small, portable, and easy to install. • Battery powered. • May be used in combination with another technology to differentiate between bicyclists and pedestrians. Limitations • Cannot be used to count bicyclists in mixed traffic. • Errors may arise because of occlusion with groups of pedestrians. • Device performance can be affected by extreme temperatures. 5.6.6 Accuracy NCHRP Project 07-19 tested two models of passive IR sensors at a total of nine locations (each model was tested at a subset of these locations). One model tested performed substan- tially better than the other model, with a correction factor of 1.037 versus 1.412 (a factor of 1.00 indicates that no correction is needed). The APD, representing the overall divergence from perfect accuracy across all data collected, was -3.1% for one product and -16.7% for the other product. The AAPD, representing the counter’s consistency, was 11.2% for one product and 33.1% for the other product. Figure 5-3 shows the testing results, combined for both products. Previous research has shown that passive IR sensors undercount pedestrians, with the rate of undercounting increasing as pedestrian volumes increase (Schneider et al. 2012). Although there may be higher error rates when ambient air temperature approaches normal body temperature, “no conclusive evidence of this increased error exists, and the error may vary among different brands of passive infrared counters” (FHWA 2013). NCHRP Project 07-19 testing did not find any evidence of increased counting error when air temperatures exceeded 90°F. Passive Infrared Summary Maximum user volume: Provides consistent results up to 600 users per hour; counts can be corrected at higher volumes. Detection zone width: <20 feet Typical count duration: Can be used for both short-term counts and permanent installations Typical equipment cost (2013): $1,000–3,000 Relative preparation cost: Medium (may require permitting) Typical installation time: <30 minutes for temporary installations, longer for permanent installations involving installing posts Typical data collector training time: <30 minutes Relative hourly cost: Low, equipment costs are spread over a large number of data-collection hours Mobility: Very good, equipment can be removed and taken to a new site

Sensor technology toolbox 89 5.6.7 Usage Passive IR counters have been tested in research projects and are commercially available. These sensors are one of the primary automated technologies in practice in the United States for pedes- trian counts. Of those practitioner survey respondents who performed counts, 22% used passive IR sensors as part of their pedestrian counting program and 19% used them as part of their bicycle counting program. 5.7 Active Infrared 5.7.1 Description Active infrared (IR) devices count pedestrians and bicyclists using an infrared beam between an emitter and a receiver located on opposite sides of a traveled way (e.g., path or sidewalk). When the beam is broken for a set period of time by an object crossing it, a detection is recorded. 5.7.2 Typical Application Active IR devices are most commonly used to collect bicycle or pedestrian screenline counts on multi-use paths. They can be installed temporarily or permanently. They cannot differenti- ate between pedestrians and bicyclists, so are best used on facilities with a single user type or in conjunction with a bicycle-only counting technology. 5.7.3 Installation Considerations The receiver and transmitter need to be installed facing each other with a clear line of sight between them, at distances up to 90 feet apart for some products (vendor recommendations may vary). The need to find suitable mounting locations on both sides of the facility may pose a chal- lenge when an active IR counter is used for a temporary count (a new post could be installed as part of a permanent installation). Avoid locations where pedestrians or bicyclists are likely to stop, linger, or congregate (e.g., doorways, bus stops, or street corners). Jones et al. (2010) found that an active IR device they tested was most accurate when oriented at a 45-degree angle to the facility. Source: NCHRP 07-19 testing. Figure 5-3. Accuracy and precision of passive infrared detectors. Active infrared counter installation at a test site. Source: Tony Hull, Toole Design Group.

90 Guidebook on pedestrian and Bicycle Volume Data collection 5.7.4 Level of Effort and Cost The level of effort is medium relative to other technologies and requires setting up the emit- ter and the receiver in appropriate locations. The cost of the equipment is high, relative to other technologies, although overall installation costs are medium, relative to other technologies. 5.7.5 Strengths and Limitations Strengths • Movable and easy to install. • Battery powered. • May be used in combination with another technology to differentiate between bicy- clists and pedestrians. • Very precise—error function is highly linear, so applying a multiplicative factor yields very accurate results. Limitations • Cannot be used for on-street monitoring. • Can count false positives from other objects (e.g. vehicles, insects, leaves, animals, and rain drops). • Errors may arise due to occlusion with groups of pedestrians or side-by-side bicyclists. • Requires mounting devices to fixed objects on each side of the trail or sidewalk. 5.7.6 Accuracy NCHRP Project 07-19 tested a single active IR device at one location. The device had fairly high accuracy and very high precision. The APD, representing the overall divergence from per- fect accuracy across all data collected, was -9.1%. The AAPD, representing the counter’s con- sistency, was 11.6%. The relationship between the automated counts and ground truth counts is almost linear, as seen in Figure 5-4. The rate of undercounting gradually increases as volumes increase. Jones et al. (2010) tested an active IR device. It was found to undercount travelers, with accu- racy rates between -12% to -18% for all travelers, and -25% to -48% for pedestrians. An inverse relationship was found between accuracy and flow. Active Infrared Summary Maximum user volume: Provides consistent results for over 600 users per hour Detection zone width: >20 feet Typical count duration: Can be used for both short-term counts and permanent installations Typical equipment cost (2013): >$3,000 Relative preparation cost: Medium (may require permitting, need to find suitable locations for both an emitter and a receiver) Typical installation time: <1 hour for temporary installations Typical data collector training time: <30 minutes Relative hourly cost: Low, equipment costs are spread over a large number of data-collection hours Mobility: Good, equipment can be removed and taken to a new site.

Sensor technology toolbox 91 5.7.7 Usage Of those practitioner survey respondents who performed counts, 13% used active IR coun- ters as part of their pedestrian counting program and 10% used them as part of their bicycle counting program. 5.8 Piezoelectric Strips 5.8.1 Description Piezoelectric materials emit an electric signal when they are physically deformed. Counters using this technology consist of two strips embedded in pavement across the traveled way. The electric signal is detected by the data logger; the order in which the two strips emit a signal provides direc- tionality, while the time interval between receiving the two strips’ signals provides speed. 5.8.2 Typical Application Piezoelectric strips are used for collecting bicycle counts at permanent count sites. They can be used on paved multi-use paths or cycle tracks. 5.8.3 Installation Considerations These counters require pavement cuts to install the piezoelectric material. Depending on the situation, considerable lead time may be required to obtain necessary permits, hire a contrac- tor, and schedule the installation. (This process may be simpler when the agency installing the counter also owns the roadway or facility.) The data logger is typically stored in a utility box next to the facility, which may require some excavation. Avoid placing the counter near intersections, to avoid overcounting bicyclists who must stop before crossing the intersection. 5.8.4 Level of Effort and Cost The level of effort is high relative to other technologies and requires careful installation. The equipment cost is medium relative to other technologies, but the overall installation cost is high relative to other technologies. Figure 5-4. Accuracy and precision of one active infrared detector. Source: NCHRP 07-19 testing. Bicyclist riding over piezoelectric strips. Source: Tony Hull, Toole Design Group.

92 Guidebook on pedestrian and Bicycle Volume Data collection 5.8.5 Strengths and Limitations Strengths • Provides speed and directionality data. • Discrete and not susceptible to tampering when embedded in pavement. • Can be battery powered or externally powered. Limitations • Cannot be used in mixed-flow traffic. • Specialized installation process. • May introduce errors with groups of bicyclists. 5.8.6 Accuracy NCHRP Project 07-19 tested one piezoelectric strip counter on a paved multi-use trail. The APD, representing the overall divergence from perfect accuracy across all data collected, was -11.4%. The AAPD, representing the counter’s consistency, was 26.6%. Figure 5-5 graphs the counter’s accuracy and precision. Piezoelectric strips have not been rigorously tested in the literature. 5.8.7 Usage Of those practitioner survey respondents who performed counts, 4 out of 115 respondents (3%) used piezoelectric strips as part of their bicycle counting program. 5.9 Radio Beams 5.9.1 Description Radio beam counters use a transmitter and receiver positioned on opposite sides of the facility. A radio signal is sent from the transmitter to receiver; when the beam is broken, a user Piezoelectric Strips Summary Maximum user volume: The one counter tested by NCHRP 07-19 did not provide precise counts at any volume, although, on average, the counts were relatively accurate. Detection zone width: <20 feet Typical count duration: Used for permanent count stations. Typical equipment cost (2013): $1,000–3,000 Relative preparation cost: High (requires permitting, pavement cuts, specialized installation) Typical installation time: >4 hours Typical data collector training time: <30 minutes Relative hourly cost: Low, equipment costs are spread over a long time period Mobility: Poor, installations are typically permanent Completed radio beam counter installation. Source: Karla Kingsley, Kittelson & Associates, Inc.

Sensor technology toolbox 93 is detected. Devices that use multiple radio frequencies can differentiate between pedestrians and bicyclists. 5.9.2 Typical Application Radio beam counters are used for screenline counts on sidewalks, pathways, and cycle tracks and can be used in both short-term and permanent counting applications. As with other beam- type technologies, they are subject to occlusion errors. 5.9.3 Installation Considerations The receiver and transmitter need to be installed facing each other with a clear line of sight between them. The multiple-frequency device tested by NCHRP Project 07-19 had a very nar- row recommended maximum separation (10 feet), which made it challenging to find locations to apply it, given that many multiple-use paths are at least 10 feet wide. The radio beam can pass through thin wood and plastic, so the devices can be hidden behind certain types of objects. The devices can be mounted on existing infrastructure or installed in a post, so the device is com- pletely hidden from sight. As with other beam-type technologies, locations where pedestrians or bicyclists are likely to linger should be avoided. 5.9.4 Level of Effort and Cost The level of effort is medium relative to other technologies and requires finding suitable loca- tions to mount a device on both sides of the facility. The equipment cost is high relative to other technologies, but the overall installation cost is medium. 5.9.5 Strengths and Limitations Strengths • Movable and easy to install. • Can be hidden in post to discourage tam- pering or theft. • Battery powered. Limitations • Errors with groups of pedestrians. • Requires mounting devices to fixed objects on each side of the trail within limited distance. Source: NCHRP Project 07-19 testing. Figure 5-5. Accuracy and precision of one piezoelectric strip device.

94 Guidebook on Pedestrian and Bicycle Volume Data Collection 5.9.6 Accuracy NCHRP Project 07-19 tested two radio beam products. Product A counted pedestrians and bicyclists separately, using two radio frequencies, while Product B counted a combined total of pedestrians and bicyclists using a single radio frequency. The APD for Product A, representing the overall divergence from perfect accuracy across all data collected, was -31.2% for the bicycle count and -26.3% for the pedestrian count. The AAPD, representing the counter’s consistency, was 72.6% for bicycles and 52.5% for pedestrians. The APD for Product B was -3.6% and the AAPD was 28.1%. Figure 5-6 graphs the counters’ accuracy and precision. Radio Beam Summary Maximum user volume: Provides consistent results up to 200 users per hour; counts can be corrected at higher volumes Detection zone width: <20 feet (single-frequency devices) <13 feet (multiple-frequency devices) Typical count duration: Can be used for both short-term counts and permanent installations Typical equipment cost (2013): >$3,000 Relative preparation cost: Medium (may require permitting, need to find suitable locations for both an emitter and a receiver) Typical installation time: <1 hour for temporary installations Typical data collector training time: <30 minutes Relative hourly cost: Low, equipment costs are spread over a long time period Mobility: Good, equipment can be removed and taken to a new site Source: NCHRP Project 07-19 testing. Figure 5-6. Accuracy and precision of radio beam detectors.

Sensor technology toolbox 95 The researchers faced significant difficulties evaluating the accuracy of the radio beam sen- sors because of a specific detail of the products being tested: namely, the counters defaulted to beginning a count immediately when initiated, rather than aggregating into bins beginning on the hour. This setting could be altered in an “advanced settings” menu, but most of the installers did not realize this. There has been no rigorous testing of radio beam technology in the literature, although one source anecdotally reports that a jurisdiction reported that it was the best technology they had used in 20 years of counting experience (ViaStrada 2009). 5.9.7 Usage Radio beam devices are commercially available in the United States. None of the project’s practitioner survey respondents who performed counts (100 who performed pedestrian counts and 115 who performed bicycle counts) had experience with this technology. 5.10 Thermal 5.10.1 Description Thermal devices generate infrared images by detecting body heat. They work similarly to passive infrared counters, but are mounted above the detection area. This positioning allows thermal devices to monitor the movement of persons and not just count the number of persons to pass the device. Thermal sensors are not affected by changes in ambient light. 5.10.2 Typical Application At the time of writing, thermal sensors had primarily been used for presence-detection appli- cations (e.g., traffic signal detectors and monitoring intrusions into restricted areas). One ven- dor was just entering the market with a device for performing bicycle and pedestrian counts. A thermal counter could perform screenline counts and could conceivably be used for pedestrian crossing counts within a defined area (e.g., a crosswalk). Thermal sensors would most likely be used for permanent count locations. 5.10.3 Installation Considerations Thermal devices require an external power source and an overhead installation location. 5.10.4 Level of Effort and Cost Unknown at the time of writing. 5.10.5 Strengths and Limitations Unknown at the time of writing. 5.10.6 Accuracy No rigorous testing has been performed on the accuracy of these devices. 5.10.7 Usage Thermal devices are becoming commercially available in the United States, but none of the respondents to this project’s practitioner survey reported experience with them.

96 Guidebook on pedestrian and Bicycle Volume Data collection 5.11 Laser Scanners 5.11.1 Description Laser scanners emit laser pulses in a range of directions and analyze the reflections of the pulses to determine characteristics of the device’s surroundings, including the presence of pedes- trians or bicyclists. Two varieties of laser scanners exist: horizontal and vertical. 5.11.2 Typical Application Non-motorized applications of laser scanners have primarily been indoors. One vendor states that the technology is best suited for locations with electrical power supplies. However, the tech- nology can be used for short-term counts on battery power. Laser scanners can collect pedestrian and bicyclist screenline counts, but cannot differentiate between the two modes. 5.11.3 Installation Considerations Permanent sites require an available electrical power supply. Horizontal scanners require loca- tions with no obstructions. Vertical scanners are mounted above the detection area. Avoid loca- tions where pedestrians or bicyclists are likely to stop, linger, or congregate (e.g., doorways, bus stops, or street corners). 5.11.4 Level of Effort and Cost Insufficient U.S. experience with outdoor applications to judge. 5.11.5 Strengths and Limitations Insufficient U.S. experience with outdoor applications to judge. 5.11.6 Accuracy Bu et al. (2007) report that laser scanners face operational difficulties in inclement weather, (e.g., rain, snow, and fog) due to interference with the laser pulses. 5.11.7 Usage Of the respondents to the project’s practitioner survey who perform counts, only 2 of 100 used it for pedestrian counting and 1 out of 115 used it for bicycle counting. 5.12 Pressure and Acoustic Pads 5.12.1 Description Pressure and acoustic pads are installed in ground, either flush with or under the surface. Pressure pads detect a change in force (i.e., weight) on the pad. Acoustic pads detect the passage of energy waves through the ground caused by feet, bicycle tires, or other wheels (FHWA 2013). 5.12.2 Typical Application Pressure and acoustic pads are primarily used to count pedestrians on unpaved trails. Pressure pads can also count bicyclists, while acoustic pads can only count pedestrians. Where pressure pads Installation of three pressure pads in a row to count side-by-side pedestrians and pedestrians traveling in opposite directions. Source: Linetop Ltd.

Sensor technology toolbox 97 are used on mixed pedestrian and bicyclist facilities, the software in the sensor can distinguish the pressure from bicyclists separate from pedestrians. These devices are most commonly used on unpaved multi-use paths and off-road trails where they can be buried and concealed. The pads require that pedestrians or bicyclists pass directly above them and are thus suited to situ- ations where pedestrians or bicyclists would normally travel single file. If users are expected to travel side-by-side, multiple pads can be placed side-by-side and linked to detect multiple users. Given the required installation, pressure and acoustic pads are typically used for long-term or permanent installations, although pressure pads for temporary indoor applications are also available on the market. 5.12.3 Installation Considerations Pads should be placed where users are expected to be moving. Locations near the start of a trail, benches, or notice boards should be avoided, because users may stop at those loca- tions. The number of pads installed should match the facility width, to the extent possible, to minimize bypass errors. Consideration should be given to travelers’ anticipated behavior and “desire lines” that show where users are traveling. For example, locations where users may cut corners or stray off the path are not ideal. Pads may be able to be installed in paved locations, but this will require the pavement to be removed and replaced. Pads are not appropriate for locations with ground freezes, because counts will typically not register in a hard frost. 5.12.4 Level of Effort and Cost The level of effort is high and requires installing the pads in the ground. Information about equipment cost was insufficient, but overall installation cost would be expected to be high relative to other counting technologies, given the need to install the pads in the ground. 5.12.5 Strengths and Limitations Strengths • Battery powered. • In-ground installation resists vandalism and theft. Limitations • Require users to pass directly above the sensor. • Most commonly used on unpaved trails. • Acoustic pads can only count pedestrians. 5.12.6 Accuracy The accuracy of these technologies has not been rigorously tested. 5.12.7 Usage The use of pressure and acoustic pads in the United States is uncommon. Of those respond- ing to this project’s practitioner survey who performed counts, none reported using pressure sensors to collect pedestrian and bicycle data. However, parks and recreation–focused agencies were not well-represented in the practitioner survey sample. The use of pressure and acoustic pads may be more common in other countries, as suggested by a review of available literature on the sensors.

98 Guidebook on pedestrian and Bicycle Volume Data collection 5.13 Magnetometers 5.13.1 Description Magnetometers detect bicycle activity through changes in the normal magnetic field as a bicy- cle’s metal parts pass by. Magnetometers are more commonly used as part of vehicle detection systems to detect the presence and movement of vehicles. While it may be possible to use existing motorized traffic magnetometers for counting bicyclists, the installation and configuration may not be optimal, and they are not designed for this purpose (FHWA 2013). 5.13.2 Typical Application Magnetometers are best suited to rural locations because the device is highly sensitive to fer- rous objects. Due to the magnetometer’s limited detection range, they are preferably installed where bicyclists will travel single file. Therefore, they are typically used to count bicyclists on rural bike paths or mountain bike trails. 5.13.3 Installation Considerations Installation requires excavating an unpaved area or removing pavement from a bicycle facility, followed by replacement. They are not appropriate for locations with ground freezes. 5.13.4 Level of Effort and Cost The level of effort is high and requires installing the device in the ground. Insufficient infor- mation was available about equipment cost, although overall installation cost would be expected to be high relative to other counting technologies, given the in-ground installation. 5.13.5 Strengths and Limitations Strengths • Battery powered. • In-ground installation resists vandalism and theft. Limitations • Relatively small detection area. • Limited application. 5.13.6 Accuracy The accuracy of this technology has not been rigorously tested. 5.13.7 Usage Magnetometers have been used to count and detect vehicles, but have not been widely applied to bicycle data collection. Magnetometers specifically designed for bicycle collection are com- mercially available in the United States. Of those responding to this project’s practitioner survey who performed counts, none reported using magnetometers to collect bicycle data. 5.14 Fiberoptic Pressure Sensors 5.14.1 Description Fiberoptic pressure sensors detect changes in the amount of light transmitted through an embedded fiberoptic cable, based on the amount of pressure (weight) applied to the cable. The sensitivity of the device can be adjusted to reflect the minimum or maximum weight desired to Source: TRAFx.

Sensor technology toolbox 99 be counted. A European vendor states that the technology can be used on mixed-traffic roadways to count bicycles separately from motor vehicles. 5.14.2 Typical Application Fiberoptic pressure sensors can be used for permanent count stations. The technology could be applied to exclusive bicycle facilities, pathways, mixed-traffic roadways, and sidewalks. 5.14.3 Installation Considerations Installation requires excavating a slot in the pavement and placing a fiberoptic cable in the slot. Avoid locations where users would be likely to congregate or linger, to avoid multiple detections of the same user. 5.14.4 Level of Effort and Cost The level of effort is high and requires installing a fiberoptic cable in the pavement and asso- ciated traffic control. As no commercial system was available in the U.S. market at the time of writing, no equipment cost information was available, although overall installation cost would be expected to be high relative to other counting technologies, given the in-ground installation. 5.14.5 Strengths and Limitations Strengths • Can classify users/vehicles based on their weight. • Can be used in mixed-flow traffic. • Discrete and not susceptible to tampering when embedded in pavement. Limitations • Specialized installation process. • Potential sources of error have not been rigorously tested. 5.14.6 Accuracy The accuracy of this technology has not been rigorously tested. 5.14.7 Usage Bicycle counters using fiberoptic pressure sensor technology are commercially available in Europe. At the time of writing, some of the components (e.g., fiberoptic cables and receivers) were available on the U.S. market, but a complete bicycle counting system was not marketed.