Water Surface Targets Recognition and Tracking Based on Improved YOLO and KCF Algorithms

Abstract

Main problems in the recognition and tracking of water surface targets include the recognition omission or error of small targets, and the bigger tracking error of occluded targets etc., This paper uses an improved YOLO v3 and KCF algorithms to obtain the accurate recognition and real-time tracking of water surface multi-target. Firstly, the water surface targets data set is established and preprocessed; then the improved YOLO v3 network based on Inception module is used to extract and identify the fine feature information of water surface targets. Next, the KCF algorithm is improved by using confidence judgment mechanism to avoid big tracking error for blocked target. Finally, combined with the data association algorithm, improved KCF can complete multi-target tracking. The test results show that the proposed recognition and tracking algorithm can recognize and track multiple targets on water surface and in the air accurately and continuously.