Pedestrian Target Tracking Algorithm on Fusion Detection

Abstract

Aiming at the problems of serious occlusion, deformation and rapid scale change in pedestrian tracking of mobile robot with vision, a pedestrian tracking algorithm with detection is proposed based on the effective convolution operators handcraft(ECO-HC), which solves the problems of target loss and inaccurate positioning caused by occlusion and background interference in the tracking process. Occlusion standard and model update threshold are set according to the peak value of confidence response. Furtherly, the position and scale of the target are corrected by using YOLO detection algorithm. The algorithm is verified on the pedestrian subset of OTB100 dataset. Experimental results show that the improved algorithm is optimal compared with other algorithms, and the overall accuracy and success rate are 93.50% and 91.80% respectively.