Towards a Rodent Tracking and Behaviour Detection System in Real Time

To analyze rodent behaviors in non-conditioned animal models is an important task that enables a researcher to elaborated conclusions about the effects in the behavior after drug application. Because the amount of data generated in the use of this kind of test, an automatized system that can record these behaviors becomes relevant. There are several proposals aiming at identifying and tracking the rodent in the open field maze, however, behavior identification is a highly desirable feature that is not included. Other works can identify behaviors, but due to high computational costs, special computers or devices are required. In this work, we propose an automatic system based on features computed by a stochastic filter that allows the development of rules to detect specific behaviors exhibited in the open field maze. We demonstrate that it is possible to track a rodent and identify behaviors in real-time (30 fps) and also in high speed (>100Hz) without the need of powerful devices or special conditions for the environment.