Granted Patent B2

US 9,716,837 · App. 14/027,681 · Granted Jul 25, 2017

Abstract

Claims (39)

1. A method for parking occupancy determination, said method comprising:

measuring changes in a field of view using template matching between templates learned from at least one original scene and templates learned from at least one current scene with motion removed, wherein said at least current scene with motion removed comprises a motion-removed current scene;

employing said changes measured in said field of view to adjust data indicative of at least one region of interest for vision-based processing thereof if said changes measured are above a pre-determined threshold so that an impact of said changes with respect to said field of view to performance is minimized; and

applying diagnostics with respect to said field of view without compensation to assist in remotely repositioning a camera associated with said field of view back to said at least one original scene,

wherein said templates comprise at least one M×N sub-image derived from said at least one original scene and said at least one current scene with said motion removed; and

said template matching is based on a normalized cross-correlation value as a similarity metric, wherein at run-time each image template among said templates slides against said motion-removed current scene and a location is determined, which yields a highest normalized cross-correlation value, said normalized cross-correlation value comprising said highest normalized cross correlation value.

2. The method of claim 1 wherein:

said templates comprise a set of interest points and at least one corresponding feature description of said at least one region of interest; and

said template matching is selected from a group of template matching comprising SIFT-matching and SURF-matching and wherein said template matching is applied in a feature space that includes said at least one corresponding feature description; and

deriving a final matched location from said template matching based on coordinates of matched interest points including at least some interest points among said set of interest points and based on a distortion/image-registration model comprising offsets, scaling, shearing rotation, and a projective transform.

3. The method of claim 1 further comprising automatically learning said templates from representative image templates of said at least one region of interest, said templates representing varying image content associated with said at least one original scene.

4. The method of claim 3 further comprising performing said template matching between said templates learned with respect to said at least one original scene and said at least one current scene with said motion removed, wherein said motion is removed to determine optimal matches at a pre-determined schedule.

5. The method of claim 4 wherein said motion is removed by averaging K frames comprising a most recent set of K frames.

6. The method of claim 4 wherein said motion is removed by background estimation using a weight sum of a previous background and a current frame of said at least one current scene with said motion removed.

7. The method of claim 6 wherein said motion is removed by a Gaussian mixture model.

8. The method of claim 6 further comprising:

tracking time of day data associated with each of said templates; and

selectively performing image registration with respect to said templates based on said time of day data with respect to said each of said templates.

9. A system for parking occupancy determination, said system comprising:

a processor; and

a non-transitory computer-usable medium embodying computer program code, said computer-usable medium capable of communicating with the processor, said computer program code comprising instructions executable by said processor and configured for:

measuring changes in a field of view using template matching between templates learned from at least one original scene and templates learned from at least one current scene with motion removed, wherein said at least current scene with motion removed comprises a motion-removed current scene;

employing said changes measured in said field of view to adjust data indicative of at least one region of interest for vision-based processing thereof if said changes measured are above a predetermined threshold so that an impact of said changes with respect to said field of view to performance is minimized; and

applying diagnostics with respect to said field of view without compensation to assist in remotely repositioning a camera associated with said field of view back to said at least one original scene,

wherein said templates comprise at least one M×N sub-image derived from said at least one original scene and said at least one current scene with said motion removed; and

said template matching is based on a normalized cross-correlation value as a similarity metric, wherein at run-time each image template among said templates slides against said motion-removed current scene and a location is determined, which yields a highest normalized cross-correlation value, said normalized cross-correlation value comprising said highest normalized cross correlation value.

10. The system of claim 9 wherein:

said templates comprise a set of interest points and at least one corresponding feature description of said at least one region of interest; and

said template matching comprises SIFT-matching and wherein said template matching is applied in a feature space that includes said at least one corresponding feature description; and

a final matched location is derived from said template matching based on coordinates of matched interest points including at least some interest points among said set of interest points and based on a distortion/image-registration model comprising offsets, scaling, shearing rotation, and a projective transform.

11. The system of claim 10 wherein said instructions are further configured for defining said at least region of interest as suitable for image registration facilitated by said distortion/image registration model performed online for every K frames rather than every frame.

12. The system of claim 10 herein said instructions are further configured for automatically learning said templates from representative image templates of said at least one region of interest, said templates representing varying image content associated with said at least one original scene.

13. The system of claim 12 wherein said automatically learning occurs offline and is updated online.

14. The system of claim 12 wherein said instructions are further configured for performing said template matching between said templates learned with respect to said at least one original scene and said at least one current scene said motion removed to determine optimal matches at a pre-determined schedule.

15. The system of claim 14 wherein said instructions are further configured for: determining an amount and characteristics of said changes in said field of view if said template matching scores above said pre-determined threshold.

16. A non-transitory processor-readable medium storing code representing instructions to cause a process for parking occupancy determination, said code comprising code to:

measure changes in a field of view using template matching between templates learned from at least one original scene and templates learned from at least one current scene with motion removed, wherein said at least current scene with motion removed comprises a motion-removed current scene;

employ said changes measured in said field of view to adjust data indicative of at least one region of interest for vision-based processing thereof if said changes measured are above a pre-determined threshold so that an impact of said changes with respect to said field of view to performance is minimized;

apply diagnostics with respect to said field of view without compensation to assist in remotely repositioning a camera associated with said field of view back to said at least one original scene, wherein each of said templates comprise at least one M×N sub-image derived from said at least one original scene and said at least one current scene with said motion removed and wherein said template matching is based on a normalized cross-correlation value as a similarity metric, wherein at run-time each image template among said templates slides against said motion-removed current scene and a location is determined, which yields a highest normalized cross-correlation value, said normalized cross-correlation value comprising said highest normalized cross correlation value.

Assignments (8)