Vehicle Tracking Method Based on Pyramid Layered Optical Flow Network
Jingya Cheng, Zhiqiang Ma, Caijilahu Bao, Wenjun Gao, Leixiao Li, Hongbin Wang · 2022
In order to solve the problem that the vehicles associated with the track cannot be uniquely identified when tracking a moving vehicle by MOT approaches in computer vision field, a moving vehicle tracking method based on pyramid hierarchical optical flow network was proposed. The Pyramid Optical Flow Net (POFN) was used to estimate the motion of moving vehicles, and the optical flow map containing the movement information of vehicles was obtained. The action refined net was designed to identify the optical flow map and estimate the more accurate position of vehicles. The motion information of the vehicle was correlated with the Intersection over Union (IOU), and the cosine similarity algorithm was used to verify the correlation of the appearance information of the vehicle boundary frame after the correlation of the motion information, so as to complete the vehicle tracking. The test results of UA-DETRAC data set show that the number of track ID switching of POFN model is less than that of current advanced tracking methods.