Adaptive Correlation Filtering Algorithm for Video Target Tracking based on Multi Feature Fusion

Abstract

In recent years, the target tracking algorithm based on the correlation filter has become a hot topic in the field of target tracking with its excellent performance in tracking precision and tracking speed. Aiming at problem of the insufficient accuracy of complex scene image and video image data, a multi feature fusion adaptive correlation filtering algorithm for video image tracking is proposed. In this work, 36 groups of color video sequences in the tracking benchmark database (OTB-2013) are utilized as samples. Firstly, histogram of oriented gradient (HOG) and color name (CN) are used to extract the two complementary features from video sequences. Then these features are used to train correlation filters and according to the complementarity of features, the response graph of two correlation filters is weighted together to effectively for tracking the image targets. After that, the confidence level of response graph and the intra-frame variation rate are calculated to dynamically adjust the learning rate and update the parameters of two correlation filters. Finally, the scale adaptive estimation algorithm is introduced to achieve scale adaptive tracking of targets. The experimental results from OTB-2013 tracking datum database show that the multi feature fusion adaptive correlation filter is suitable for complex scene video image data and the accuracy and speed of automatic tracking is improved.