The Effect of Object Features on Multiple Object Tracking and Identification

Tracking of multiple objects is challenging for computer vision system under certain circumstance. We investigated this problem with human observers. In our experiment, observers were asked to track multiple moving items as well as to maintain their identities. We found that the capacity of maintaining multiple moving object identities of human is about three to four items, and uniqueness improves the general tracking and ID performance. It also showed that observers’ capacity of ID task was dependent on feature type, and suggests that a less resource-demanding process of identity-related feature would lead to more effective improvement on tracking. These results provide some indications for the design of computer vision system which involves human monitoring, and suggest that creating a featural space to map the identity of multiple objects may aid the automatic object tracking.