Research On Computer Vision-Based Fish Movement Behavior Monitoring System

Biological monitoring techniques for water quality are comprehensively used in water environment quality assessment. The basic principle of biomonitoring is that using the reaction of biological individual, population and community to environmental pollution to expound the pollution state of the environment, and provide monitoring data for the environmental quality assessment. The movement behavior of the hydrobios is the most important index of the water safety warning system based on biomonitoring, because the change of water biology's movement behavior can reflect the change of the water quality directly and real-timely. So, how to get the features of hydrobios' movement behavior is the basic and important part of the study of biomonitoring-based water quality monitoring and safety warning system.This study used fish as the indicator organisms of biomonitoring for water quality monitoring, and mainly study the application of computer vision in fish movement behavior monitoring. The main purpose of this paper is that using computer vision techniques to acquire the features of fish movement behavior, including fish swimming trace, speed, acceleration, and so on. The specific content of this study including: real-time detecting fish objects, tracking multiple fish targets and developing the software of fish movement behavior monitoring system. The works of this paper are sdtudied as follows:First, study the methods of real-time fish detecting. The traditional average background model can not update the background frame accurately, so this paper proposes an improved adaptive background model which takes account of frame difference to update the background frame. The new background model divides the background frame image into four different regions according to the frame difference, and updates each part with unique factor separately. It makes the background frame updated more quickly and accurately. And meanwhile, an online auto threshold method is used to binary the background difference and frame difference images of video sequence. It narrows the field of gray levels in which Otsu method finds the very threshold value corresponding to the Maximal Variance Between-Class of the image to save the computation time.Second, study the particle filter based multiple fish targets tracking algorithm. Particle filter is efficient to estimate the non-linear and non-gaussian bayesian process' posterior state. This paper studies the theory of particle filtering and applies it to track multiple fish targets in video sequence. The very similar appearance and randomly movement of fish are the most challenges to track multiple fish. This paper defines the single fish's movement state's dynamic model, and also models the interaction actions between multiple fish targets. Further more, a multi-object interactive observation model is proposed to avoid the limitations of independent particle filter-based multi-object tracking. It can get rid of the false observation as much as possible when several fish targets get closed and move interactively between each other. This proposed method can track multiple fish targets efficiently.Last, study the development of the software of computer vision-based fish movement behavior monitoring system. In this paper, study the architecture of the computer vision-based fish movement behavior monitoring system, and develop the prototype software of the system. Further more, the software analyses the fish's swimming trace obtained by tracking, and establish the swimming speed and acceleration models and so on. It provides data for water quality monitoring and water safety warning.