Abstract: Traditional kernel correlation filter (KCF) algorithm uses HOG features to obtain target information, which is not robust to non-rigid targets and is prone to target tracking drift. This paper proposes a target tracking algorithm that combines optical flow features and saliency detection to suppress tracking drift. The algorithm integrated optical flow features into the expression of multi-channel features to increase the change information of the position and posture of the moving target. At the same time, re-detection and adjustment of drifting targets were performed through the saliency detection position to suppress tracking drift and improve tracking accuracy. Experimental results show that the proposed algorithm can still perform robust visual target tracking in complex scenes.