Multi-target tracking algorithm based on deep learning

Abstract With the continuous development of deep learning in multi-target tracking, use convolutional neural network for feature extraction has replaced traditional method, but accuracy target tracking needs to be improved. In order further improve a new algorithm based on RFB is proposed this paper. The mainly divided into three parts: detection, and tracking. detection part, CenterNet was selected as accuracy. RFBNET combined with pedestrian re-recognition strengthen capability. DeepSORT used multi - Experimental results MOT16 data set show that more effective than other methods.