Research On Swimming Tracking Of Medaka Fish Based On Computer Vision

Computer vision,also known as machine vision,is a new research field.It is a technology that simulate the human visual system and process the external information by using camera and computer technology.Computer vision includes different research directions,among which object detection and recognition,moving target tracking,3d reconstruction are the most popular.In recent years,with the rapid development of computer technology,computer vision technology has been applied to many traditional industries and achieved a good practical effect in operation.Fish behavior research is the foundation of aquaculture development,efficient fishing and fishery resources sustainably utilizated.Fish swimming behavior is the reaction for the external environment.The date of fish swimming behavior is an important source of information to investigate fish life habits.To some extent,fish swimming trajectory reflects the adaptability of fish to their environment.Speed and acceleration data are important indicators of fish swimming ability in a specific environment.Traditional fish swimming behavior monitoring mainly by human operating,generally cost much labor and inefficiency.Therefore,how to acquire fish swimming behavior data efficiently and stably has became the main subject of fish behavior research.In this paper,the fish target in video sequence was detected and tracked based on computer technology,and a method of locating and extracting fish target using Gaussian mixture model,morphological processing and contour analysis model was proposed.This paper upgraded the kernelized correlation filter tracking algorithm,improved the Camshift tracking algorithm,and constructed the fish swimming monitoring system platform,so this paper has important significance for studying the living habits of fish in specific environments.In this paper,medaka is selected as the research object,and the main work includes the following three parts:(1)This paper summarizes the background and current situation of fish behavior research,and reviews the fish detection and tracking algorithms.This paper analyzes the principle and existing problems of the classical motion detection and tracking algorithm,and provides an important basis for later motion detection and tracking algorithm.(2)The accurate fish target location extraction method based on gaussian mixture model,morphological operation and contour analysis model was studied.Firstly,the video image sequence is grayscale processed,and then the image is filtered and removed by gaussian template.Finally,the accurate fish swimming targets are extracted by gaussian mixture model,morphological processing and contour analysis model.The enhanced kernelized correlation filter tracking algorithm is studied to track the fish.Firstly,the information of the current frame and the previous frame is trained to obtain the multi-core correlation filter,and then the correlation operation is carried out with the new input frame.Finally,the response value obtained by the operation can be used as the predicted tracking result,and the maximum response value is the tracking target position.The results show that the enhanced kernelized correlation filter tracking algorithm can track fish swimming behavior stably in real time,and the extracted fish swimming trajectory is highly consistent with the real-time record trajectory.Based on the traditional Camshift tracking algorithm,a method combining Camshift tracking algorithm and Kalman algorithm is proposed.At the same time,BP neural network is introduced to improve the accuracy of Kalman algorithm prediction.The results show that the improved Camshift algorithm can track the fish stably and effectively under the condition of fast movement or severe occlusion,with good anti-interference ability and robustness.(3)Using C plus plus as development language,Open CV as computer vision library,Visual studio2015,Qt5.9.0 as development framework,fish swimming monitoring system is independently developed with a detailed introduction of the whole framework,The video image acquisition module included in the system is also described.Video image processing module and fish tracking module implementation,as well as a variety of implementation of the algorithm,to meet the requirements of the different environment.At the same time,it can locate and track the swimming position of fish,display and output swimming position,speed and track data of fish in real time,And the generated data are stored to provide data support for the subsequent analysis of fish swimming behavior.The innovative achievements of this paper are mainly reflected in the following aspects:(1)By using Gaussian mixture background modeling and contour analysis model,the fish swimming target is located and extracted accurately.(2)Combined the traditional kernelized correlation filter tracking algorithm with the enhanced kernelized correlation filter tracking algorithm,fish swimming trajectory,position and speed data can be obtained in real time.At the same time,the swimming territories of different sizes of fish in the tank were analyzed.On the basis of Camshift algorithm,Camshift is combined with Kalman algorithm,and BP neural network is introduced to improve the accuracy of Kalman algorithm.(3)Designed and completed medaka fish monitoring system based on computer vision,realized the image processing of fish swimming video and the tracking of fish swimming,and acquired the data of fish swimming motion parameters in real time,which provided a strong support for the systematic study of fish swimming behavior.