In multicamera surveillance systems, tracking the same person across multiple cameras is an important technique. It is also desirable to recognize the individuals who have been previously observed with a single-camera monitor. This paper addresses this problem by recognizing the same individual in different tracks. The method that represents an object image using a bag of features has been commonly used in image retrieval and classification. In this paper, that approach is adapted for people image description, and support vector machines are employed for high classification performance. To get more reliable matches and support supervised learning in online operation, we propose a decision scheme to distinguish previously unseen individuals from recurrences so that the new classes can be automatically labeled. On this basis, an online recognition framework that applies incremental learning is also presented. We get promising results from the evaluation with more than 200 tracks of 70 different people.
Kun Liu, Kun Liu,
Jie Yang, Jie Yang,
"Online recognition of people recurrences with bag-of-features representation and automatic new-class labeling," Optical Engineering 49(1), 017203 (1 January 2010). https://doi.org/10.1117/1.3281668