An Appearance Based Approach for Human and Object Tracking
MARYLAND UNIV COLLEGE PARK LANGUAGE AND MEDIA PROCESSING LAB
Pagination or Media Count:
We have implemented a system for tracking humans and detecting human-object interactions. Persistent tracking of humans and objects in a video sequence is an important task in surveillance application. Pose and illumination variations, occlusion, appearance and disappearance of humans in the scene, etc... are some of the challenges one has to face. We present an appearance based approach to the problem. A combination of correlogram and histogram information is used to model object and human color distributions. Humans and objects are detected using a background subtraction algorithm. The models are built on the fly and used to track them on a frame by frame basis. The system is able to detect when humans merge into groups and segment them during occlusion. Identities are preserved during all the sequence, even if a person enters and leaves the scene. The system is also able to detect when a person deposits or removes and object from the scene. In the first case the models are used to track the object retroactively in time. In the second case the objects are tracked for the rest of the sequence. The model is able to overcome common deformations as well as many situations involving occlusion. Furthermore, it is easy to update. We assume a static camera and focus on compressed images taken in an indoor environment. The results show that this is a powerful processing technique providing important information to algorithms performing higher level analysis such as activity recognition, where human-object interactions play an important role.
- Numerical Mathematics
- Optical Detection and Detectors