System and method for tracking movement of objects in a scene
First Claim
1. A method for tracking movement of objects in a scene from a stream of video frames, each video frame comprising image information of the scene at an instant of time, the method comprising steps of:
- (a) generating from the image information region clusters which are hypotheses of objects to be tracked in the scene, wherein multiple region clusters are generated for each object to be tracked in the scene so as to provide multiple candidates of the same object;
(b) generating a confidence value for each region that represents a likelihood that the region cluster represents an object to be tracked, determining a correspondence between region clusters in consecutive video frames; and
(c) constructing tracks of objects through the scene over time based upon the correspondence determined between region clusters in consecutive video frames, wherein multiple tracks are constructed for each object based on correspondence between video frames determined for each of the plurality of region clusters associated with each object to be tracked.
20 Assignments
0 Petitions
Accused Products
Abstract
A system and method for tracking movement of objects, such as people, through a scene. A background image comprising image information representing a background of the scene is generated. Image information for a video frame is compared with image information of the background image to generate regions in a video frame which potentially represent objects to be tracked in the scene. Region clusters are formed by combinations of regions if such combination satisfy predetermined criteria, and the region clusters are hypotheses of objects to be tracked. Each region cluster is evaluated to determine whether each represents an object to be tracked, based on object model information and correspondence to region clusters in a prior video frame. A confidence value is generated for each region cluster that represents a likelihood that the region represents an object to be tracked. A correspondence is determined between region clusters in consecutive video frames. Tracks of objects through the scene over time are constructed based upon the correspondence determined between region clusters in consecutive video frames.
404 Citations
16 Claims
-
1. A method for tracking movement of objects in a scene from a stream of video frames, each video frame comprising image information of the scene at an instant of time, the method comprising steps of:
-
(a) generating from the image information region clusters which are hypotheses of objects to be tracked in the scene, wherein multiple region clusters are generated for each object to be tracked in the scene so as to provide multiple candidates of the same object;
(b) generating a confidence value for each region that represents a likelihood that the region cluster represents an object to be tracked, determining a correspondence between region clusters in consecutive video frames; and
(c) constructing tracks of objects through the scene over time based upon the correspondence determined between region clusters in consecutive video frames, wherein multiple tracks are constructed for each object based on correspondence between video frames determined for each of the plurality of region clusters associated with each object to be tracked. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8)
(d)(1) forming a first correspondence graph comprising a plurality of object nodes each representing a region cluster and a plurality of tracks, each track comprising an ordered sequence of object nodes in consecutive video frames that represents a track segment of an object in the scene; and
(d)(2) forming a second correspondence graph comprising a plurality of nodes and a plurality of tracks, each node in the second correspondence graph corresponding to at least one track in the first correspondence graph, and each track comprising an ordered sequence of nodes in the second correspondence graph representing a track of an object through the scene.
-
-
9. A system for tracking movement of objects in a scene comprising:
-
(a) a video camera positioned to monitor the scene and generate video signals representing activity within the scene;
(b) a frame grabber coupled to the video camera to generate a stream of video frames from the video signal, each video frame comprising image information of the scene at an instant of time;
(c) a processor coupled to the frame grabber, the processor being programmed to;
(1) generate from the image information region clusters for a video frame which are hypotheses of objects to be tracked in the scene, wherein multiple region clusters are generated for each object to be tracked in the scene so as to provide multiple candidates of the same object;
(2) generating a confidence value for each region that represents a likelihood that the region cluster represents an object to be tracked, determine a correspondence between region clusters in consecutive video frames; and
(3) construct tracks of objects through the scene over time based upon the correspondence determined between region clusters in consecutive video frames, wherein multiple tracks are constructed for each object based on correspondence between video frames determined for each of the plurality of region clusters associated with each object to be tracked. - View Dependent Claims (10, 11, 12, 13, 14, 15, 16)
(i) forming a first correspondence graph comprising a plurality of object nodes representing region clusters and a plurality of tracks, each track comprising an ordered sequence of object nodes in consecutive video frames that represents a track segment of an object in the scene; and
(ii) forming a second correspondence graph comprising a plurality of nodes and a plurality of tracks, each node in the second correspondence graph corresponding to at least one track in the first correspondence graph, and each track comprising an ordered sequence of nodes in the second correspondence graph representing a track of an object through the scene.
-
Specification