Method for visual-based recognition of an object
First Claim
Patent Images
1. A method for visual-based recognition of an object, said method comprising:
- receiving digital depth data for at least a pixel of an image of an object, which is not required to be inside of a subject, said depth data comprising information relating to a distance from a visual sensor to a portion of said object shown at said pixel, said visual sensor comprising an emitter and sensor of light, wherein said light is selected from the group of electromagnetic radiation consisting of visible light, infrared light, and ultraviolet light and wherein said receiving of said depth data does not require special behavior from one of said object and said subject;
generating a two dimensional plan-view image based in part on said depth data, wherein said generating includes generating said plan-view image as if said object was viewed at an axis normal to ground level from above and wherein generating other view images based on different orientations of said object other than at said axis normal to ground level from above is not required;
extracting a plan-view template from said plan-view image, wherein at least a portion of said plan-view image is transformed; and
processing said plan-view template at a classifier, that is executing on a computer system, to assign a class to said plan-view template, wherein said classifier is trained to make a decision according to pre-configured parameters determined at least in part based on said class of said plan-view template.
1 Assignment
0 Petitions
Accused Products
Abstract
A method for visual-based recognition of objects is described. Depth data for at least a pixel of an image of the object is received, the depth data comprising information relating to a distance from visual sensor to a portion of the object visible at the pixel. At least one plan-view image is generated based on the depth data. At least one plan-view template is extracted from the plan-view image. The plan-view template is processed by at least one classifier, wherein the classifiers are trained to make a decision according to pre-configured parameters.
-
Citations
40 Claims
-
1. A method for visual-based recognition of an object, said method comprising:
-
receiving digital depth data for at least a pixel of an image of an object, which is not required to be inside of a subject, said depth data comprising information relating to a distance from a visual sensor to a portion of said object shown at said pixel, said visual sensor comprising an emitter and sensor of light, wherein said light is selected from the group of electromagnetic radiation consisting of visible light, infrared light, and ultraviolet light and wherein said receiving of said depth data does not require special behavior from one of said object and said subject; generating a two dimensional plan-view image based in part on said depth data, wherein said generating includes generating said plan-view image as if said object was viewed at an axis normal to ground level from above and wherein generating other view images based on different orientations of said object other than at said axis normal to ground level from above is not required; extracting a plan-view template from said plan-view image, wherein at least a portion of said plan-view image is transformed; and processing said plan-view template at a classifier, that is executing on a computer system, to assign a class to said plan-view template, wherein said classifier is trained to make a decision according to pre-configured parameters determined at least in part based on said class of said plan-view template. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22)
-
-
23. A visual-based recognition system comprising:
-
a visual sensor for capturing depth data for at least a pixel of an image of an object, which is not required to be inside of a subject, said depth data comprising information relating to a distance from said visual sensor to a portion of said object visible at said pixel, said visual sensor comprising an emitter and sensor of light, wherein said light is selected from the group of electromagnetic radiation consisting of visible light, infrared light, and ultraviolet light and wherein said capturing of said depth data does not require special behavior from one of said object and said subject; a plan-view image generator for generating a two dimensional plan-view image based on said depth data, wherein said generating of said plan-view image includes generating said plan-view image as if said object was viewed at an axis normal to ground level from above and wherein generating other view images based on different orientations of said object other than at said axis normal to ground level from above is not required; a plan-view template generator for generating a plan-view template based on said plan-view image; and a classifier for making a decision concerning recognition of said object, wherein said classifier is trained to make said decision according to pre-configured parameters that were determined at least in part based on a class assigned to said plan-view template. - View Dependent Claims (24, 25, 26, 27, 28, 29, 30, 31)
-
-
32. A method for visual-based recognition of an object representative in an image, said method comprising:
-
generating a three-dimensional point cloud based on digital depth data for at least a pixel of an image of said object, which is not required to be inside of a subject, said depth data comprising information relating to a distance from a visual sensor to a portion of said object visible at said pixel, said visual sensor comprising an emitter and sensor of light, wherein said light is selected from the group of electromagnetic radiation consisting of visible light, infrared light, and ultraviolet light, said three-dimensional point cloud representing a foreground surface visible to said visual sensor and wherein a pixel of said three-dimensional point cloud comprises a three-dimensional coordinate and wherein said generating of said three-dimensional point cloud does not require special behavior from one of said object and said subject; partitioning said three-dimensional point cloud into a plurality of vertically oriented bins; mapping at least a portion of points of said vertically oriented bins into at least one said plan-view image based on said three-dimensional coordinates, wherein said plan-view image is a two-dimensional representation of said three-dimensional point cloud comprising at least one pixel corresponding to at least one vertically oriented bin of said plurality of vertically oriented bins, wherein said mapping includes generating said plan-view image as if said object was viewed at an axis normal to ground level from above; and processing said plan-view image at a classifier, that is executing on a computer system, wherein said classifier is trained to make a decision according to pre-configured parameters and wherein said pre-configured parameters were determined based at least in part on a class assigned to a plan-view template that was extracted from said plan-view image by transforming at least a portion of said plan-view image, said classifier does not require other view images based on different orientations than at said axis normal to ground level from above of said object in order to make said decision. - View Dependent Claims (33, 34, 35, 36, 37, 38, 39, 40)
-
Specification