Monitoring mice social behaviors is extremely important for neurobehavioral analysis. State-of-the-art monitoring systems still require human handling for phenotype characterization with high cost and low standardization. Mice tracking and identity preservation is the first step for phenotyping. This paper focuses on a new computer-vision-based automated tracking system able to identify mice and keep their identities frame by frame, laying the groundwork for automatic social behavior analysis. Our system achieves more than 80% accuracy on metal ear tags identification on one-minute long videos recorded at 30 fps.