Robust Movement-Specific Vehicle Counting at Crowded Intersections

Zhongji Liu, Wei Zhang, Xu Gao, Hao Meng, Xiao Tan, Xiaoxing Zhu, Zhan Xue, Xiaoqing Ye, Hongwu Zhang, Shilei Wen, Errui Ding; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 614-615

With the demands of intelligent traffic, vehicle counting has become a vital problem, which can be used to mitigate traffic congestion and elevate the efficiency of the traffic light. Traditional vehicle counting problems focus on counting vehicles in a single frame or consecutive frames. Nevertheless, they are not expected to count vehicles by movements of interest (MOI), which can be pre-defined by all possible states of vehicles, combining different lanes and directions. In this paper, we mainly focus on movement-specific vehicle counting problems. A detection-tracking-counting (DTC) framework is applied, which detects and tracks objects in the region of interest (ROI), then counts those tracked trajectories by movements. To be specific, we propose the detection augmentation method and the Mahalanobis distance smoothness method to improve the multi-object tracking performance. For vehicle counting, a shape-based movement assignment method is carefully designed to categorize each trajectory by movements. Experiments are conducted on both the AICity 2020 Track-1 Dataset and the Vehicle-Track Dataset, which is built in this paper. Experimental results show the effectiveness and efficiency of our method.