Research On Key Technologies For Video Online Monitoring Of Fish Behavior

Biological monitoring technique is to assess the quality and state of the environment in abiological system by biological assessment techniques and methods, using the indicators such asphysical characteristics and behavior reaction of water-based creatures. It is widely used in watersafety warning and water monitoring system and it provides theoretical basis and informationbasis for the assessment system of water safety from a biological viewpoint. Water biologicalmonitoring results, especially for fish behavior, can reflect the influence and the degree of harmwhich is from environmental quality and change. Therefore, it is one of the most direct andeffective methods of water quality monitoring. So, How to rapidly extract and analyze theseindicators of biological characteristics and online monitor these water-based creatures are keypoints of the research for online monitoring of water quality.This thesis considers fish behavior in water as an indicator organism, and mainly discussesthe application of computer vision technology in the field of online monitoring of fish behavior.The research target is to obtain the feature information of fish behavior for biological waterquality monitoring. The specific studies include: moving object detection method based onminimum cross entropy, single fish tracking method based on Camshift algorithm, multi-fishtracking method based on particle filter algorithm, design and implementation of onlinemonitoring system of fish behavior. The main research work is summarized as follows:First, research on moving object detection based on minimum cross entropy. In the stage ofimage binarization when selecting the threshold, this thesis introduces the concept ofgray-gradient co-occurrence matrix model and approximately selects the optimal threshold bygenetic algorithms, and then assesses the quality of image segmentation using the measurementof uniformity.Second, research on single target tracking method based on Camshift algorithm. Based ontraditional Camshift algorithm, this thesis proposes a novel method, which can locate the targetautomatically. This method can solve the issues of tracking lost caused by the fast motion oftarget, so that it can improve the accuracy and stability of the single-target tracking. Third, research on multi-target tracking method based on particle filter algorithm. In theframework of particle filter, this thesis combines inter-frame relational matrix and classifier andproposes a method of multi-target particle filter tracking. The method can sufficiently make useof the information source of detector and classifier, and realize the detecting and tracking formulti-fish. The method improves the robustness and is suitable for the scene of online real-timeapplication which is a combination of relationship matrix and classifier.Forth, design and implementation of video online monitoring system of fish behavior.Based on the research on the detecting and tracking algorithms of moving objects, the thesisdesigns and implements a preliminary online monitoring system of fish behavior.