Mobile robot following pedestrian based on high confidence update KCF.

The real-time tracking of human targets by mobile robot was prone to problems such as scale variations, occlusion and deformation of targets. Aiming at above problems, an improved kernel correlation filtering algorithm (OURS algorithm) based on high-confidence update strategy of model and scale was proposed. The method of one-dimensional fast discriminative scale space was used to solve the problem that the kernel correlation filtering algorithm can't deal with the scale variation. A new confidence measurement method was proposed to judge the detection results of OURS algorithm model. When the target was completely occluded, it switched to Kalman filter algorithm to predict the position of the target. The confidence degree of the scale detection was judged to avoid the wrong scale update of tracking box. Through the video sequence simulation experiment and real environment test, the target tracking effect of the OURS algorithm was verified and compared with that of kernel correlation filter(KCF) algorithm and fast discriminative scale space tracking (fDSST) algorithm. The results show that the proposed OURS algorithm has the precision of 0.758, and the success rate of 0.711, and the target tracking effect of OURS algorithm was obviously improved compared with KCF, fDSST algorithms. The tracking box can still accurately frame the target after the scale variation, occlusion, deformation of pedestrian target, and other factors. The pedestrian tracking system of mobile robot based on image servo control strategy can adjust the linear velocity and angular velocity to maintain a stable following according to the position of the pedestrian. [ABSTRACT FROM AUTHOR]