Fish detection and movement tracking

Fish Detection and Tracking is an important step in studying oceanography, especially for forecasting changes in the quality of water and the increasing or decreasing number of fish in a population. In this paper, combination of Gaussian Mixture Model and Frame-Differencing algorithm (CGMMFD) is proposed to improve tracking performance in different scenarios. Also, four other techniques, namely Mean Background, Gaussian Mixture Model, Mean Shift Tracking and Particle Filter are also investigated. In this study, we use the self-built database with some typical tracking situations such as appearance of illusions, different swimming velocities of the fish and qualities of water. Mean square error and Variance are used to assess the performance of each technique for different scenarios. The experimental results indicate that our proposed algorithm gives higher tracking accuracy. While other techniques have difficulties to track the fish location or the fish centroid in some certain scenarios, the proposed algorithm can perform well in different situations.