An Algorithm for Tracking Multiple Fish Based on Biological Water Quality Monitoring

Abstract

Abnormal water quality will increase the occlusion rate among fish schools, which causes difficulties in fish detection and tracking. In order to solve this problem, a multiple fish tracking algorithm for red snapper is proposed in this paper. In the detection stage, we use the Otsu adaptive segmentation algorithm to extract fish targets based on the background subtraction method, following which the fish tracking feature parameters can be obtained based on the fish geometric features. In the tracking stage, the Kalman filter is employed to first estimate the motion state, and then the cost function is constructed from the position of the fish body, target area, and the direction information. Finally, fish school tracking is realized by the interframe relationship matrix. We applied several tracking methods with various schemes to experimental videos of swimming fish schools in different environments. The experimental results show that the proposed tracking algorithm exhibits improved performance and robustness. INDEX TERMS Data association, feature detection, multi-object tracking, water quality monitoring.

Application of computer vision in fish intelligent feeding system—A review

“…As long as the length of the fish is calculated, the weight of the fish can be roughly inferred according to the growth model formula (Irigoyen‐Arredondo et al., 2016; Lamprakis & Kallianiotis, 2003). The use of cameras and other equipment is an important part of fish monitoring and behaviour analysis (Zhao et al., 2016, 2019). We can only use machines to analyse how they move in relation to the state of fish.…”

“…The flow of the red snapper multi‐fish tracking algorithm is as follows (Zhao et al., 2019). In the detection stage, they use the Otsu adaptive segmentation algorithm to extract fish targets based on the background subtraction method, following which the fish tracking feature parameters can be obtained based on the fish geometric features.…”

Zebrafish tracking using YOLOv2 and Kalman filter

“…To accurately estimate the movement of the fish, the Kalman filter was used to perform the tracking task. Because the movement of the fish is considered to be a uniform linear motion from one frame to the other, the system can be approximated as a linear dynamic model 9,10,17 . The algorithm settings start with the status vector of the fish's location and orientation.…”

Multiple Object Tracking for Occluded Particles

“…We calculated the corresponding average error detection per frame (AEDF), accuracy rate and recall rate. The accuracy rate and recall rate were calculated using the expressions shown as (5) and (6), respectively [19]. Where TP is the total number of correctly detected particles, FP represents the number of false particles, and FN is the count of undetected particles that are mistaken as noise or background.…”

“…The cost of the Hungarian algorithm was the loss between predicted states of the current frame particles and the measured states. Kalman filter [19,20] was used to estimate the states. For the proposed and global shortest path algorithm, the states were not estimated.…”