Research on Pedestrian Tracking Algorithm in Industrial Scene Based on Improved ByteTrack
Ling Wang, Junxu Bai, Peng Wang, Yane Bai · 2023
ByteTrack algorithm is one of the algorithms with good tracking effect at present, but it is still challenging for problems such as missing detection and noise. This paper proposes a pedestrian tracking algorithm for industrial scene based on ByteTrack algorithm and names as G-ByteTrack. Firstly, using the SG Kalman filter to replace the Kalman filter, SG Kalman filter can smooth the sudden change of noise, assist the prediction of target position, and improve the tracking accuracy; Then, using GSI interpolation method to make up some missing trajectory gap and further improve the accuracy of tracking trajectory. On the self-made industrial pedestrian tracking dataset, the comparison experiment between G-ByteTrack algorithm and ByteTrack algorithm shows that MOTA is increased by 2.69%, IDF1 is increased by 1.97%, and the ID switching issues is improved by 8%.