This paper describes a method of object tracking that can be applied to real-time scene analysis applications. The proposed method combines spectral blob analysis combined with zone-based detection. Objects are identified by their color signatures and then tracked in the camera's field of view. The field of view is mapped into zones that represent logical separations of the scene, such as lanes of traffic. The method was implemented with the CMUcam2, which is a low cost sensor used in the experiments. The results demonstrate that the method developed can track multiple color objects at rates as high as 15 frames per second. The system also stores a behavioral history of the object motion within the scene, allows for the detection of occurrences such as lane changes.
Download Full PDF Version (Non-Commercial Use)