Motion detection and classification: ultra-fast road user detection

Abstract With the emerge of intelligent and connected transportation systems, driver perception on-board safety systems could be extended with roadside camera units. Computer vision can utilised to detect road users, conveying their presence vehicles that cannot perceive them. However, accurate object detection algorithms are typically computationally heavy, depending on delay-prone cloud computation or expensive local hardware. Similar problems faced in many applications, which users detected a camera. We propose utilising Motion Detection Classification (MoDeCla) for user detection. The approach is lightweight capable running real-time an inexpensive single-board computer. To validate applicability MoDeCla benchmark was carried out manually labelled data gathered from surveillance cameras overseeing urban areas Espoo, Finland. Separate datasets were during winter summer, enabling comparison detectors significantly different weather conditions. Compared state-of-the-art detectors, performed order magnitude faster, yet achieved similar accuracy. most impactful deficiency errors bounding box placement. Car headlights long dark shadows found especially difficult motion detection, caused incorrect boxes. Future improvements also required separately detecting overlapping users.