Enhancing the quality of infrared-based automatic pedestrian sensor data by nonparametric statistical method

With the advent of intelligent transportation systems in the past several decades, sensors are being extensively used to detect and count vehicle movements. The use of similar sensing technologies to detect pedestrian movements, however, is relatively new. Pedestrian counts are essential for decision making in pedestrian facility planning, signal timing, and pedestrian safety modeling. Conventional methods such as manual counting and videotaping can hardly satisfy the requirements of programs for the long-term collection of pedestrian data. Advances in sensing technologies have increased the ability to automate pedestrian data collection with the use of infrared sensors. However, the quality of the data from infrared sensors is still a problem; several field studies have shown that these types of sensors do not always perform perfectly. Field tests conducted in this study and by other research teams showed that infrared sensors usually counted significantly fewer pedestrians than the actual number. Thus, the quality of the data from infrared sensors needs to be enhanced. This paper proposes a nonparametric statistical method to calibrate raw sensor data to achieve this goal. A bivariate bootstrap sampling procedure was used to obtain correction factors for new counts instead of the traditional regression-based approach. Two case studies were used to test the validation of the proposed calibration procedure. Test results showed that the proposed procedure could improve the quality of the sensor data by reducing the discrepancy between sensor counts and ground truth (true) data. The transferability of the calibration procedure was also verified through the case studies.