Edge-Based Traffic Flow Data Collection Method Using Onboard Monocular Camera

Abstract

Traffic data collection is the fundamental step in most applications of intelligent transportation systems (ITS). Recently, traffic data collection methods have become more robust and diversified, yet still have some limitations in their flexibility and coverage. Onboard monocular cameras have considerable potential to be turned into cost-effective moving traffic sensors combining the low cost and ego-vehicles’ high mobility. Existing studies have explored the feasibility of onboard cameras for scene understanding, etc. However, few studies have been conducted to utilize onboard monocular cameras for traffic flow data collection. To this end, this paper puts forward a method using the onboard monocular camera to collect traffic data. The basic structure is composed of a you-only-look-once (YOLO) model and spatial transformer network (STN) to detect vehicles in real-time. Then the traffic flow parameters are computed via fundamental optic and traffic flow theories. The experiment results show its reliability and similar sensing accuracy with inductive loop detectors on the road segment detection. In addition, the STN-YOLO model has a higher vehicle detection accuracy than the original YOLO model under complicated conditions.