Infrared Face Part Tracking Based on Correlation Filtering

Abstract

Compared with RGB images, infrared images have simple scene structure and little background interference, to achieve more accurate target detection. Therefore, target tracking for infrared thermal imaging has also become the focus of research. In order to achieve stable tracking of infrared face under complex background and multi-interference conditions, we propose a tracking algorithm using correlation filtering based on adaptive scales. In order to extract effective features from infrared face, this paper uses a tracking framework based on correlation filters and an adaptive scale mechanism to mark 12 groups of infrared video sequences under different conditions, and track infrared face parts. Compared with the traditional algorithm, the average tracking success rate of this method is significantly improved in complex background. It can effectively track infrared targets under such factors as partial occlusion, motion blur, fast motion, illumination changes, background clutter, and scale changes.