Video Analytics in Smart Transportation for the AIC'18 Challenge

Ming-Ching Chang, Yi Wei, Nenghui Song, Siwei Lyu; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 61-68

With the fast advancements of AICity and omnipresent street cameras, smart transportation can benefit greatly from actionable insights derived from video analytics. We participate the NVIDIA AICity Challenge 2018 in all three tracks of challenges. In Track 1 challenge, we demonstrate automatic traffic flow analysis using the detection and tracking of vehicles with robust speed estimation. In Track 2 challenge, we develop a reliable anomaly detection pipeline that can recognize abnormal incidences including stalled vehicles and crashes with precise locations and time segments. In Track 3 challenge, we present an early result of vehicle re-identification using deep triplet-loss features that matches vehicles across 4 cameras in 15+ hours of videos. All developed methods are evaluated and compared against 30 contesting methods from 70 registered teams on the real-world challenge videos.