Infrared Moving-point Target Semi-Automatic Labeling Algorithm Based on Target Enhancement and Visual Tracking

Infrared video data annotation has the problems of low efficiency and poor quality. In this paper, a semi-automatic labeling method for moving point targets in infrared sequence images is proposed based on target enhancement and visual tracking to solve it. First, infrared sequence images in a continuous period of time were registered and fused to enhance the target features. Second, a visual tracking algorithm was utilized to locate the fused features efficiently and automatically. Lastly, a saliency map was obtained through phase spectrum reconstruction, and the exact coordinates of a target were obtained. During automatic annotation, the difference between the annotation results of adjacent frames was used to select key frames, which enabled the annotators to locate the image frames that had errors and manually annotated them quickly. The results of the experiments showed that the algorithm significantly reduced the participation of annotators and effectively solved the problems of long period and poor quality assurance in data annotation.