Research on Fish recognition algorithm based on Machine Learning algorithm
Download as PDF
DOI: 10.23977/iset2021.029
Author(s)
Yan Yixuan, Zhang Qichen, Shi Xiaohui
Corresponding Author
Yan Yixuan
ABSTRACT
The study of fish behavior with the help of computer vision technology has gradually become a hot topic, this technology simulates the principle of biological vision, and obtains dynamic target parameter information by processing collected pictures or videos, in order to achieve the purpose of monitoring and analysis of fish swimming behavior. Using the traditional Vibe algorithm to monitor the swimming behavior of fish will consume a large number of video frames to eliminate ghosts, and the detection results of moving targets under the dynamic background of water surface ripples are not accurate, so an improved Vibe algorithm is proposed. In view of the fact that a large number of video frames are consumed to eliminate ghosts, a hierarchical traversal search algorithm is proposed to mark and count the target image to adaptively adjust the background update probability to quickly eliminate ghosts. In view of the fact that the detection results of moving targets in the dynamic background with water surface ripples are not accurate enough, a method based on LBP and HSV to remove water surface ripples is proposed to improve the detection accuracy.
KEYWORDS
Moving target detection, Vibe algorithm, ghost removal, dynamic background