Target Tracking Algorithm Combining Improved GMS and Correlation Filtering

Abstract

Aiming at the problem that traditional target tracking algorithms cannot detect and track specific targets in the real environment (such as airports) due to the given initial frame target position, this article proposes a target tracking algorithm that combines improved Grid-based Motion Statistics (GMS) matching algorithm and correlation filtering tracking algorithm. First of all, for the problem that GMS cannot adapt to the detection of specific targets in realistic monitoring environment, a template size expansion method is proposed to disperse the centrally gathered feature points and Random Sample Consensus (RANSAC) is introduced to remove outliers in the center of similar areas. Second, the initial template of the tracking algorithm is the target detected by the improved GMS method, and it is used to extract features to train the correlation filter to determine the target position of the video sequence. Finally, two groups of comparative experiments show that not only the improved GMS algorithm has better target detection performance, but also the fusion algorithm has better reliability and robustness for target tracking.