Accurate counting of pedestrians is one of the great challenges in traffic data collection. Conventional data collection methods such as manual counting hardly satisfy the requirements of long-term pedestrian studies. Technological advancements accelerated the development of automatic pedestrian counting devices. Practices show that none of automatic counter performsperfectly. There is need to improve the automatic counting performance. Therefore, this paper attempts to propose a novel calibration procedure to estimate more reliable counts using the raw sensor outputs. Unlike previous studies, this paper focuses on revealing the internal relationship between pedestrian arrival patterns and counter performance. Lab experiments and field tests are conducted to establish and validate the statistical relationships between actual counts, sensor counts and arrival patterns. Statistical tests illustrate that there are no statistically significant differences between the calibrated and actual counts.
Abstract