Posture tracking of multiple individual fish for behavioral monitoring with visual sensors

A novel scheme for the behavioral monitoring of multiple fish is presented based on measurements of the fish posture. Image identification of the fish head and tail is proposed by examining the gray level appearance features. A simplified posture of fish was calculated from its head and centroid. Multiple individual tracking was developed by incorporating the fish posture and global nearest neighbor. The proposed tracking scheme was tested with 2-5 individual Zebra fish (Cyprinidae, Danio rerio) under laboratory conditions. The experimental results confirmed the stability of the identification of a fish head and tail and a robust tracking performance was achieved. The proposed method could accurately record the movement trajectories and posture information of each individual. The tracking errors were only 0.278%-1.572% for 2-5 individual fish while the posture measurement error rates were less than 0.16%. Furthermore, the movement patterns of the fish, e.g., angular variations and distribution of the movement angle, were investigated from the behavioral data, which revealed the advantages of the proposed posture measurement for behavioral studies. The aggregation behavior of multiple fish was also analyzed to reveal the interaction patterns of the individuals in response to the spatial formation of the neighbors.