A tracking method of multiple moving objects for animal watch sensor systems — Measures for random mobility and the problem with connection and separation

Behavior analysis of birds is useful to detect avian-flu infected birds in large cages. In this study, we developed an automatic avian flu detection system using camera images. The chicken paths are extracted from an image sequence captured with a monocular camera using particle filters. However, the system often fails to track multiple targets because of random mobility and the apparent connection and separation. As countermeasures, we devised a method to perform clustering of moving object areas as preprocessing. Moreover, we performed some experiments to confirm the availability of our method, and achieved robust tracking with particle filters.