The behavior of fish in the vicinity of fishing gear and in an aquaculture tank should be analyzed to understand their responses to materials such as nets, tank walls, and other physical objects. To monitor the 3D trajectory of target fish in recorded videos automatically and thereby reduce the workload of scientists, we used a particle filter as a state-space model, which is an image processing technique, to analyze video. The behavior of a single chub mackerel of fork length of 0.34 m was monitored using stereo vision cameras in an experimental fish tank. The fish body in the recorded video was binarized using threshold and background subtraction methods of image processing. Then, a segmented image of the fish body including color information was generated by obtaining the product of the raw image and binarized image. The segmented fish body was automatically tracked in the video recorded by the stereo camera using a particle filter. The 3D trajectory of the fish was calculated using direct linear transformation methods based on the 2D trajectories in the stereo image. The trajectory was smoothed using the Kalman filter to minimize measurement error in estimated fish position. Thus, using the proposed method based on image processing, we automatically obtained the 3D trajectory of the target fish. Although the wave effect in field observation remains an issue in applying this method to fisheries research, it can be put to immediate use in tank experiments.