Abstract:Mice interaction recognition has been widely employed in animal observation experiments since it can provide useful biological signals for researchers to identify the social behaviors of animal models. Rather than manual observation, automatic observation systems based on computer vision can be used to detect and track the mice behaviors dynamically, which has become popular in mice interaction recognition. In order to enhance the accuracy of recognition, a novel mice interaction recognition method by using machine learning is proposed in this paper. A new elliptical model is developed to fit every mouse and to extract its motion and shape features. For selecting the optimal feature, we investigate the influence of different features on the result. We improve the average recognition result on the novel dataset RatSI.

A novel mice interaction recognition method by using machine learning