초록 ▼

A method and/or system for tracking objects, such as humans, over a wide area (that is, over an area that is delineated by a large spatial domain and/or a long-duration temporal domain) is provided. Such tracking is facilitated by processing, in real-time, near real-time or otherwise contemporaneous

A method and/or system for tracking objects, such as humans, over a wide area (that is, over an area that is delineated by a large spatial domain and/or a long-duration temporal domain) is provided. Such tracking is facilitated by processing, in real-time, near real-time or otherwise contemporaneous with receiving, images captured by each of a plurality or network of slightly overlapping stereo sensors, such as stereo cameras. The method includes and the apparatus is adapted for obtaining a plurality of local-track segments, wherein the plurality of local-track segments correspond to an object captured in images taken by a respective plurality of stereo sensors; and combining the local-track segments to form a global track.

대표청구항 ▼

What is claimed is: 1. A method for tracking objects over a wide area, the method comprising: using a processor to generate a plurality of sets of local-track segments, wherein the plurality of sets of local-track segments respectively correspond to a plurality of objects captured in images taken b

What is claimed is: 1. A method for tracking objects over a wide area, the method comprising: using a processor to generate a plurality of sets of local-track segments, wherein the plurality of sets of local-track segments respectively correspond to a plurality of objects captured in images taken by a respective plurality of stereo sensors, wherein generating each of the plurality of sets of local-track segments is based on generating, for each of the plurality of objects, a non-parametric probabilistic-shape model for such object; generating, for each of the plurality of objects, a part-based appearance representation for such object, wherein the part-based appearance representation includes generating adaptive and fixed image templates using global representations; generating, for each of the plurality of objects, a ground-based-depth representation for such object; and iteratively estimating ownership and motion parameters of each of the non-parametric probabilistic-shape model, the part-based appearance representation, and the ground-based-depth representation using an Expectation Maximalization algorithm; and designating at least one of the plurality of objects as a designated set of objects; combining, for each of the designated objects, its set of local-track segments along with space and time constraints to form a designated global track, wherein the space and time constraints are formed from ground-based, space-time cues. 2. The method of claim 1 wherein generating from each of the images the plurality of sets of local-track segments comprises using an object-representation process on each of the images to detect the plurality of objects and to form the plurality of sets of local-track segments. 3. The method of claim 2, wherein using an object-representation process on each of the images comprises integrating multiple cues of the plurality of objects. 4. The method of claim 3, wherein integrating multiple cues of the plurality of objects comprises integrating, for each of the plurality of objects, position, shape, appearance and depth cues of such object. 5. The method of claim 1, wherein generating, for each of the plurality objects, a part-based appearance representation comprises generating, for each of the plurality of objects, a color histogram of appearance. 6. The method of claim 1, wherein generating from each of the images the plurality of sets of local-track segments comprises: using fast-object segmentation to differentiate each of the objects from another so that the corresponding set of local-track segments are associated with each of the plurality of objects. 7. The method of claim 6, wherein using fast-object segmentation to differentiate each of the objects comprises using an efficient-mode-seeking algorithm on a foreground-ground-accumulation map one each of the images. 8. The method of claim 1, wherein obtaining a plurality of sets of local-track segments, designating at least one of the plurality of objects, and combining, for each of the designated objects, its set of local-track segments to form a designated global track are performed in real-time, near real-time or otherwise contemporaneous with obtaining the images captured by the plurality of stereo sensors. 9. The method of claim 1, wherein the space and time constraints comprise object states selected from the group of objects states consisting of ground location and ground velocity. 10. The method of claim 1, wherein the space and time constraints comprise object states, and further comprising: modeling the object states using a Gaussian variable, and estimated the object states using a Kalman filter operated in any of a forwards and/or backwards direction.