NJDOT needs to collect accurate pedestrian related information in a cost effective way. According to the RFP issued by NJDOT, there are key gaps for pedestrian planning and mobility including the “number of pedestrians using any given sidewalk, path, crosswalk, or other pedestrian facilities”. The lack of such data is in turn clearly one of the one of the most significant barriers to the development of safety conscious transportation plans that includes pedestrians as well as vehicles. The same RFP states two important types of information needed for reliable decisionmaking:

1. better understanding of pedestrian behavior,
2. more accurate and complete inventory of pedestrian flow rates.

In the past, pedestrian count information was generally collected manually. However, since the manual collection of accurate pedestrian counts can be quite expensive and time-consuming, this approach is used sporadically and as a result does not yield comprehensive data from which to make informed policy and planning decisions. In fact, because of extensive time and labor requirements of manual data collection, which might also be relatively inaccurate, reliable pedestrian flow information is most of the time not available to the planners and decision makers.

Emerging sensor technologies accelerated the shift toward automatic pedestrian counting methods to acquire reliable long-term data for transportation design, planning, and safety studies. Although a number of commercial pedestrian sensors are available, their accuracy under different pedestrian traffic flow conditions is still questionable. Moreover, it is difficult to assess the suitability of different sensors for different locations. Some sensors claimed to be more accurate are substantially more expensive. Ease of deployment, power requirements, and long-term deployment issues all play an important role in the selection of sensors. This study attempts to shed light on the understanding of field performance of two commercially available automatic pedestrian sensors by performing rigorous comparisons—namely, a passive infrared counter by EcoCounter and a thermal sensor by TrafSys. A major innovation of this study was to simultaneously deploy the two relatively different sensor technologies—thermal and infrared sensors—under the same experimental conditions to compare their performances. To achieve this in a statistically robust manner, pairwise tests were conducted at trails and intersections with different pedestrian flow levels and characteristics. Statistically significant differences in terms of accuracy were found. The thermal sensor was found to produce less error than EcoCounter, which significantly undercounted pedestrians at intersections. This result was expected since EcoCounter is recommended for trail settings. The results also demonstrated the variability of both sensors given different deployment conditions. A calibration procedure for the EcoCounter data was also presented.