A real-time zebrafish tracking based on fish body analysis

Zebrafish as a model organism has been widely chosen in biological and ecological sciences. The behavior of zebrafish represents the effects of disturbances and stress from the environment. Recent, advanced computer vision techniques have been changing traditional biological and ecological research and providing abundant behavioral information on the experimental targets (e.g., zebrafish). In this work, we propose a real-time zebrafish tracking method based on the fish body analysis from the fish contour images. The advantage of analyzing contour images is that could reduce noises or errors from the image degradation, such as image blurring caused by fast motion. We also presented an automatic fish parameter estimation scheme providing easy use to the users. Moreover, the determination of fish occlusion is implemented based on fish parameter estimation. In this work, fish skeleton and distance transform are utilized to detect the fish head and tail point from the single fish and overlapped fish images. And the orientation of the fish head and tail is estimated according to their gradient direction. Fish head and its direction are chosen as the major features for tracking multiple fishes. A cost matrix is built according to the fish head and direction information. Accordingly, individual fish trajectories are assigned by solving the linear assignment. The proposed method was tested on a 5 fish video clip. The results showed that the correct tracking rate reaches 0.998 and the correct identification rate after occlusion is 0.94. The run-time of the proposed method is better than 30 fps and satisfies the requirement of real-time monitoring. The proposed method presents a solution for long-term multiple fish observation applications.