An Automated Fish Behavior Observation And Measure System Based On Video

The purpose of studying fish behavior is to find out fish migration routes and improve capture efficiency, which gives an important foundation for the protection of fishery resources. A large number of data is necessary, such as stress response,cluster,migration and other measured data, but getting these data is a time-consuming process. Fish behavior is recorded with the form of video generally in the laboratory study. How to extract the data of fish behavior efficiently from the video has been a major problem of automated fish tracking. We successfully used the advanced technology of military radar tracking maneuvering multiple targets in fish behavior research. An automated fish 3D information tracking system based on video is a new method that is put forward in this article, which has a great significance to further study of fish behavior.Hemigrammus rhodostomus is taken as the target fish of this study. By using a single camera and a waterproof-mirror, we extract the data of fish behavior automatically by fish behavior observation and measure system which is divided into four parts: (1)Distortion calibration of single camera system;(2)Transfer formula between image coordinate to world coordinate;(3)The automated tracking algorithm of fish movement;(4)The automated output of fish behavior 2D and 3D data.The main research contents are:(1)Distortion calibration of single camera system: draw a sealed waterproof paper with 8×10 uniform grids before pasting it on a glass board and then put it into the experimental aquaria. Taking photos with Canon ISU1200 IS camera while changing the angle of the board from time to time. Extract more than 20 images in different angles and input them into the tool-box of Camera Calibration by Matlab. It shows that: pixel error becomes smaller when the focus lens of the camera is in the front of the aquaria. Adjusting the distance between the camera and the aquaria to 1.5 m, then the result shows pixel error is about 0.1.(2)Transfer formula between image coordinate to world coordinate: after background subtraction, we consider the centroid of the fish as its 2D position. The lower left vertex on the front of the aquaria was taken as the origin of the coordinate system and we assume X-axis to the right while Y-axis to the inside and Z-axis to the upwardness. As the camera tilted slightly during the experiment, the shape of the aquaria in the images changed. So according to the theory of free transferred disposal, we have to correct the deformation of images. After scaling conversion from pixel to centimeter, X-axis and Z-axis of 3D fish coordinate can be read right from 2D image coordinates. Y-axis of 3D fish coordinate can be calculated from Z-axis of fish reflection in the mirror.(3) The automated tracking algorithm of fish movement: combine Interacting Multiple Model Algorithm with Joint Probabilistic Data Association. We successfully used the advanced technology of military radar tracking maneuvering multiple targets in fish behavior research, and an automated fish 3D information tracking system based on video is put forward in this article.3D tracking of the fish can be divided into three parts: firstly define the motion estimation model of the fish: as the complicated motion of fish, an single motion model can not describe it exactly, we use Interacting Multiple Model Algorithm; secondly, the connection of each frame of the same fish: linking the location in last frame with the location in the next frame of the same fish in order to get movement trajectory of the fish; thirdly, Kalman filter: predict and correct fish tracking by using Kalman filter.(4) The automated output of fish behavior data: the purpose of establish the fish automated observing system is to obtain the motion parameters of target fish, especially the fish's location,speed,direction and so on. In the process of calculating fish 3D coordinates, we use VC program to release the tracking of fish behavior automatically and output 3D data of fish behavior in the same time. The data was saved in txt and xls format.The result shows that:We successfully used IMM and JPDA algorithm in the automated fish behavior observation and measure system, and it works quite well. The combine of IMM and JPDA algorithm can deal with the key issues during fish tracking system. With much more fish number, JPDA algorithm computation becomes much more difficult. Calculation load increased exponentially under maneuvering multiple targets environment. According to the difficult computation issue, we combine NN algorithm and JPDA algorithm to solve the problem, which means the first choice is NN algorithm and when NN algorithm is not able to work things out we use JPDA instead.During fish tracking process, phenomenon of missing happens when the light becomes weak, we have to lower the detect value near the forecast position and make further research to improve the rate of target detection. When two fish block each other for seconds, JPDA algorithm is used until they are apart.