Target Tracking Algorithm of Similarity Detection Based on Correlation Filtering

Abstract

Tracking-Learning-Detection (TLD) algorithm is a better method in the study of a continuous single target tracking algorithm. However, the optical flow method used in TLD is an ideal model based on three assumptions, errors are prone to occur in the change of object scale, and the errors will accumulate gradually. In view of the limitations of TLD tracker, this paper proposes a fusion algorithm of Sale Adaptive with Multiple Features (SAMF) tracker and Kernelized Correlation Filter (KCF) tracker. Similarity is calculated by tracking results, and the detection module switching is judged by similarity. The target tracking frame is adjusted by combining the two results. The fusion algorithm can stably and well output tracking results in complex situations such as blur, fast moving, object occlusion, illumination change and has good robustness, which is suitable for long-term target tracking.