Long-term Tracking Algorithm Based on Dimensionality Reduction and Re-Detection

Aiming at the tracking failure caused by target occlusion,scale change and illumination change in long-term target tracking,a long-term target tracking algorithm combined with dimensionality reduction and re-detection is proposed.Based on translation and scale estimation from a long-term correlation tracking algorithm,the dimensionality reduction strategy of principal component analysis was adopted to reduce the computational burden.At the same time,a high confidence sample set was established.When the target was occluded or lost for a long time,the online classification detector and the best-buddies similarity matching were started through the adaptive threshold to relocate the target position,and a balanced model updating strategy was used to update the template.The experimental results of quantitative and qualitative evaluation on some sequences of standard data sets such as OTB-2015 show that the average distance accuracy of the algorithm in this paper is 95.4%,the average overlap success rate is 89.2%,and the average tracking speed is 23.68 frames/s.Moreover,the algorithm performs well in scenes such as occlusion,scale change and illumination change,and can effectively achieve long-term target tracking.