A Fast and Robust Multiple Individuals Tracking Algorithm Based on Artificial Neural Networks

To understand the complex collective phenomena caused by interaction mechanism between individuals, researchers are paying more and more attention to the data-driven approaches. The crucial data is the trajectory of each individual in a group, it is necessary to detect and track the individuals automatically. Unfortunately, due to various shape, complex motion pattern and frequent occlusions of fish, it is difficult to automatically track fish and acquire trajectories. Existing methods work with only one specie and have all sorts of limitations such as high resolution, high quality video, longer detecting and recognition time or steady background. These methods will have low performance and cost more time for low-quality videos. To deal with above problem, we developed a fast and robust multiple individuals tracking algorithm based on the combination of artificial neural network and background subtraction. We use neural network and background division for target recognition. We use hungarian matching algorithm for inter-frame matching, and construct benefit matrix by combining individual velocity, area, acceleration and other factors. Finally, we test our algorithm for different experimental videos, the results show that our method has good performance in terms of detecting and tracking speed, trajectory integrity and usability. At the same time, our algorithm also supports the trajectory tracking of ants, fruit flies and other organisms.