Method and apparatus for pedestrian detection
First Claim
1. A method of vision processing comprising:
- producing a depth map of a scene proximate a vehicle;
comparing the depth map to a pedestrian template;
identifying a match between the depth map and the pedestrian template, wherein objects having inverse eccentricities less than a threshold are not identified as a match; and
detecting the presence of a pedestrian.
2 Assignments
0 Petitions
Accused Products
Abstract
A vehicle vision system that identifies pedestrians located proximate a vehicle. The system comprises a sensor array that produces imagery that is processed to generate disparity images. A depth map can be produced from the disparity images. Either the depth map or the disparity images are processed and compared to pedestrian templates. A pedestrian list is produced from the results of that comparison. The pedestrian list is processed to eliminate false pedestrians as determined by inverse eccentricity. The pedestrians are then tracked, and if a collision is possible, that information is provided to the driver and/or to an automated collision avoidance or damage or injury mitigation system.
98 Citations
40 Claims
-
1. A method of vision processing comprising:
-
producing a depth map of a scene proximate a vehicle;
comparing the depth map to a pedestrian template;
identifying a match between the depth map and the pedestrian template, wherein objects having inverse eccentricities less than a threshold are not identified as a match; and
detecting the presence of a pedestrian. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A method of vision processing comprising:
-
producing a multi-resolution disparity image of a scene proximate a vehicle;
comparing the multi-resolution disparity image to a pedestrian template;
identifying a match between the multi-resolution disparity image and the pedestrian template, wherein objects having inverse eccentricities less than a threshold are not identified as a match; and
detecting the presence of a pedestrian. - View Dependent Claims (18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32)
-
-
33. Apparatus for performing vision processing comprising:
-
a stereo image preprocessor for producing a multi-resolution disparity image;
a depth map generator for processing the multi-resolution disparity image to form a depth map; and
a target processor for comparing the depth map to a plurality of pedestrian templates to identify a match between at least one pedestrian template within the plurality of pedestrian templates and the depth map, wherein the target processor does not identify objects having inverse eccentricities less than a threshold. - View Dependent Claims (34, 35)
-
-
36. A pedestrian detection system comprising:
-
a platform;
a stereo camera pair attached to said platform;
a stereo image preprocessor for producing a multi-resolution disparity image from said stereo camera pair;
a depth map generator for processing the multi-resolution disparity image disparity image to form a depth map; and
a pedestrian processor having a pedestrian template database, said pedestrian processor for comparing the depth map to a plurality of pedestrian templates in said pedestrian template database to identify a match between at least one pedestrian template and the depth map, wherein the pedestrian processor does not identify objects having inverse eccentricities less than a threshold as a pedestrian.
-
-
37. A pedestrian detection system comprising:
-
a vehicle;
a stereo camera pair attached to said vehicle;
a stereo image preprocessor for producing a multi-resolution disparity image from said stereo camera pair; and
a pedestrian processor having a pedestrian template database, said pedestrian processor for comparing the multi-resolution disparity image to a plurality of pedestrian templates in said pedestrian template database to identify a match between at least one pedestrian template and the multi-resolution disparity image, wherein the pedestrian processor does not identify objects having inverse eccentricities less than 0.4 as a match as a pedestrian.
-
-
38. A computer readable medium for storing a computer program that directs a computer to:
-
produce a depth map of a scene proximate a platform;
compare the depth map to a pedestrian template;
identify a match between the depth map and the pedestrian template, wherein objects having inverse eccentricities less than a threshold are not identified as a match; and
detect the presence of a pedestrian. - View Dependent Claims (39, 40)
-
Specification