It is an intuitive and efficient method to monitor the water quality using the biological characteristics of aquatic organisms. The paper studies a vision-based perceptive framework for fish motion, of which some modules are studied, such as video data capture, moving object detection and multiple object tracking and so on. A multi-object tracking using particle filter with interacting observing model is proposed, and some related kinematical data, i.e., velocity and acceleration, are defined and analyzed to represent the real-time fish activity. The experimental results show that it is efficient and accurate.