Tracker Toolkit

This software can track multiple moving objects within a video stream simultaneously, use visual features to aid in the tracking, and initiate tracks based on object detection in a subregion. A simple programmatic interface allows plugging into larger image chain modeling suites. It extracts unique visual features for aid in tracking and later analysis, and includes sub-functionality for extracting visual features about an object identified within an image frame. Tracker Toolkit utilizes a feature extraction algorithm to tag each object with metadata features about its size, shape, color, and movement. Its functionality is independent of the scale of objects within a scene. The only assumption made on the tracked objects is that they move. There are no constraints on size within the scene, shape, or type of movement. The Tracker Toolkit is also capable of following an arbitrary number of objects in the same scene, identifying and propagating the track of each object from frame to frame. Target objects may be specified for tracking beforehand, or may be dynamically discovered within a tripwire region. Initialization of the Tracker Toolkit algorithm includes two steps: Initializing the data structures for tracked target objects, including targets preselected for tracking; and initializing the tripwire region. If no tripwire region is desired, this step is skipped. The tripwire region is an area within the frames that is always checked for new objects, and all new objects discovered within the region will be tracked until lost (by leaving the frame, stopping, or blending in to the background).