DETECTED OBJECT TRACKER FOR A VIDEO ANALYTICS SYSTEM
Techniques are disclosed which provide a detected object tracker for a video analytics system. As disclosed, the detected object tracker provides a robust foreground object tracking component for a video analytics system which allow other components of the video analytics system to more accurately evaluate the behavior of a given object (as well as to learn to identify different instances or occurrences of the same object) over time. More generally, techniques are disclosed for identifying what pixels of successive video frames depict the same foreground object. Logic implementing certain functions of the detected object tracker can be executed on either a conventional processor (e.g., a CPU) or a hardware acceleration processing device (e.g., a GPU), allowing multiple camera feeds to be evaluated in parallel.
1 . A computer-implemented method for tracking foreground objects depicted in a scene, the method comprising:
identifying a first plurality of regions of pixels classified as depicting scene foreground in a current video frame of the scene;
classifying one or more of the regions of pixels in the first plurality as depicting either a known foreground object or a discovered foreground object by matching a geometry associated with each of the first plurality of regions of pixels against a geometry associated with a second plurality of regions of pixels, wherein the second plurality of regions of pixels correspond to regions of pixels classified as depicting scene foreground in a previous video frame of the scene.
2 . The method of claim 1 , further comprising:
reclassifying at least a first one of the discovered foreground objects as depicting a known foreground object by matching a geometry associated with first discovered foreground object against a geometry associated with a third plurality of regions of pixels.
3 . The method of claim 2 , wherein the third plurality of regions of pixels corresponds to a set of missing foreground objects previously tracked in the scene.
4 . The method of claim 1 , further comprising:
identifying a first foreground object previously tracked in the scene, wherein the first foreground object did not have an associated geometry matching the geometry of one of the first plurality of regions;
generating a plurality of locations in the current frame to search for the first foreground object;
at each generated location in the current frame, comparing appearance values of pixels of the first foreground object with appearance values of pixels of the current frame to determine a measure of similarity; and
upon determining that the measure of similarity generated for a first one of the generated locations exceeds a threshold, determining an associated geometry for the first foreground object at the first location and classifying the first foreground object as one of the known foreground objects.
5 . The method of claim 3 , wherein determining the associated geometry for the first foreground object comprises determining a size and orientation of an ellipse to bound the first foreground object.
6 . The method of claim 3 , further comprising, extending a trajectory of the first foreground object based on the appearance values of the pixels and the associated geometry of the first known foreground object.
7 . The method of claim 1 , wherein the geometry associated with each of the first plurality of regions and the second plurality of regions comprises an ellipse bounding the respective region.
8 . The method of claim 1 , further comprising, extending a trajectory of each foreground object classified as depicting one of the known foreground objects based on the appearance values of the pixels and based on the associated geometry of the corresponding known foreground object.
9 . A computer-readable storage medium containing a program, which when executed on a processor, performs an operation for tracking foreground objects depicted in a scene, the operation comprising:
identifying a first plurality of regions of pixels classified as depicting scene foreground in a current video frame of the scene;
classifying one or more of the regions of pixels in the first plurality as depicting either a known foreground object or a discovered foreground object by matching a geometry associated with each of the first plurality of regions of pixels against a geometry associated with a second plurality of regions of pixels, wherein the second plurality of regions of pixels correspond to regions of pixels classified as depicting scene foreground in a previous video frame of the scene.
10 . The computer-readable storage medium of claim 9 , wherein the operation further comprises:
reclassifying at least a first one of the discovered foreground objects as depicting a known foreground object by matching a geometry associated with first discovered foreground object against a geometry associated with a third plurality of regions of pixels.
11 . The computer-readable storage medium of claim 10 , wherein the third plurality of regions of pixels corresponds to a set of missing foreground objects previously tracked in the scene.
12 . The computer-readable storage medium of claim 9 , wherein the operation further comprises:
identifying a first foreground object previously tracked in the scene, wherein the first foreground object did not have an associated geometry matching the geometry of one of the first plurality of regions;
generating a plurality of locations in the current frame to search for the first foreground object;
at each generated location in the current frame, comparing appearance values of pixels of the first foreground object with appearance values of pixels of the current frame to determine a measure of similarity; and
upon determining that the measure of similarity generated for a first one of the generated locations exceeds a threshold, determining an associated geometry for the first foreground object at the first location and classifying the first foreground object as one of the known foreground objects.
13 . The computer-readable storage medium of claim 12 , wherein determining the associated geometry for the first foreground object comprises determining a size and orientation of an ellipse to bound the first foreground object.
14 . The computer-readable storage medium of claim 12 , wherein the operation further comprises, extending a trajectory of the first foreground object based on the appearance values of the pixels and based on the associated geometry of the first known foreground object.
15 . The computer-readable storage medium of claim 9 , wherein the geometry associated with each of the first plurality of regions and the second plurality of regions comprises an ellipse bounding the respective region.
16 . The computer-readable storage medium of claim 9 , wherein the operation further comprises, extending a trajectory of each foreground object classified as depicting one of the known foreground objects based on the appearance values of the pixels and the associated geometry of the corresponding known foreground object.
17 . A system, comprising:
a video input source configured to provide a sequence of video frames, each depicting a scene;
a central processing unit (CPU); and
a memory containing a program, which, when executed on the CPU is configured to perform an operation for tracking foreground objects depicted in a scene, the operation comprising:
identifying a first plurality of regions of pixels classified as depicting scene foreground in a current video frame of the scene, and
classifying one or more of the regions of pixels in the first plurality as depicting either a known foreground object or a discovered foreground object by matching a geometry associated with each of the first plurality of regions of pixels against a geometry associated with a second plurality of regions of pixels, wherein the second plurality of regions of pixels correspond to regions of pixels classified as depicting scene foreground in a previous video frame of the scene.
18 . The system of claim 17 , wherein the operation further comprises:
reclassifying at least a first one of the discovered foreground objects as depicting a known foreground object by matching a geometry associated with first discovered foreground object against a geometry associated with a third plurality of regions of pixels.
19 . The system of claim 18 , wherein the third plurality of regions of pixels corresponds to a set of missing foreground objects previously tracked in the scene.
20 . The system of claim 17 , wherein the operation further comprises:
identifying a first foreground object previously tracked in the scene, wherein the first foreground object did not have an associated geometry matching the geometry of one of the first plurality of regions;
generating a plurality of locations in the current frame to search for the first foreground object;
at each generated location in the current frame, comparing appearance values of pixels of the first foreground object with appearance values of pixels of the current frame to determine a measure of similarity; and
upon determining that the measure of similarity generated for a first one of the generated locations exceeds a threshold, determining an associated geometry for the first foreground object at the first location and classifying the first foreground object as one of the known foreground objects.
21 . The system of claim 20 , wherein determining the associated geometry for the first foreground object comprises determining a size and orientation of an ellipse to bound the first foreground object.
22 . The system of claim 20 , wherein the operation further comprises, extending a trajectory of the first foreground object based on the appearance values of the pixels and based on the associated geometry of the first known foreground object.
23 . The system of claim 17 , wherein the geometry associated with each of the first plurality of regions and the second plurality of regions comprises an ellipse bounding the respective region.
24 . The system of claim 17 , wherein the operation further comprises, extending a trajectory of each foreground object classified as depicting one of the known foreground objects based on the appearance values of the pixels and the associated geometry of the corresponding known foreground object.