AIC2018 Report: Traffic Surveillance Research

Traffic surveillance and management technologies are some of the most intriguing aspects of smart city applications. In this paper, we investigate and present the methods for vehicle detections, tracking, speed estimation and anomaly detection for NVIDIA AI City Challenge 2018 (AIC2018). We applied Mask-RCNN and deep-sort for vehicle detection and tracking in track 1, and optical flow based method in track 2. In track 1, we achieve 100% detection rate and 7.97 mile/hour estimation error for speed estimation.