Detection and tracking from real-time video streams
The main objective of this project is to demonstrate that the interactive parts model, that was initially applied to segmentation and recognition of one-dimensional objects such as handwritten words, can be successfully extended to segmentation and detection of two-dimensional objects such as vehicles.
Toward this end we built a specialized environment for image acquisition and image processing that allows users to interactively modify the parameters of the recognition system while the system is running. The environment allows the user to stop the video at any particular frame and analyze it and then either continue
the video or advance it to the next frame. The bi-directional exchange of information between the system and the user is facilitated through a specially constructed dialog window (Filter Properties). This window allows the user, among other things, to: a) edit and adjust the camera parameters, b) select the mode for system operation by choosing between a tracker or a detection mode, c) select the method for calculating the background image, and d) label the objects/features, or verify whether the system correctly labeled them, and collect various statistics.
