Adaptive Traffic Light Control System with Vision Based Traffic Flow Sensing Method

This invention relates to a system installed at an intersection that captures image feeds of vehicles, accelerates vehicle detection using a custom coprocessor implemented on the FPGA fabric of a programmable system on chip (PSoC), and hence adaptively controls traffic signals at an intersection. The invention further includes methods for detecting vehicles from a video feed, tracking vehicles, measuring weighted vehicle flow, and hence calculating signal timings.

The methods for detecting vehicles, tracking vehicles, measuring weighted vehicle flow are demonstrated to generalize well with remarkable accuracy in challenging lighting and weather conditions, when evaluated in day, night and rain in multiple real world intersections. The method for calculating traffic signal timings from weighted vehicle flow measurement is demonstrated to be stable, allowing 100-300 more vehicles per hour through the intersection and reducing average waiting time of vehicles by 3-5 seconds per vehicle per cycle compared to the static timing values used in the real world when evaluated on a real world intersection modelled with real world traffic flow data in VISSIM traffic simulation software.