Granted Patent B2

US 8,917,904 · App. 12/108,581 · Granted Dec 23, 2014

Inventors: Wende Zhang (Shelby Township, MI); Mark J. Wolski (Warren, MI)

Assignee: GM Global Technology Operations LLC

G06K9/00791

Abstract

Claims (65)

1. A method for vehicle clear path detection using a camera comprising:

imaging a ground area in front of said vehicle with said camera to produce a ground image; and

analyzing said ground image by grouping all individual objects upon ground together as part of an overall uniform limitation and subtracting the overall uniform limitation from said ground image to formulate a clear path free of objects limiting travel of said vehicle including

iteratively identifying a component patch of said ground image, each iteration corresponding to a different component patch identified,

extracting, during each iteration, a feature from respective ones of said component patches, and

classifying each of said component patches based upon respective ones of said features, said classifying comprising

determining, during each iteration for each of said component patches, a respective patch clear path likelihood describing a fractional confidence that said respective feature is indicative of a clear path free of all objects limiting travel of said vehicle, wherein each respective patch clear path likelihood is determined by

analyzing raw features a-priori during a training stage using a training set of images to obtain distinguishing features, and

assigning the respective clear path likelihood of said respective feature based on the distinguishing features obtained during the training stage,

comparing, during each iteration for each of said component patches, said respective patch clear path likelihood to a threshold clear path confidence, said threshold clear path confidence is a predetermined value constant for all determined patch clear path likelihoods,

designating respective ones of said component patches as clear if respective ones of said patch clear path likelihoods are greater than said threshold clear path confidence, and

designating respective ones of said component patches as not clear if respective ones of said patch clear path likelihoods are not greater than said threshold clear path confidence, wherein each component patch designated as not clear comprises part of the overall uniform limitation subtracted from said ground image to formulate said clear path.

2. The method of claim 1 , wherein extracting a feature comprises feature recognition based on filtering said component patch.

3. The method of claim 1 , wherein said threshold clear path confidence is substantially 0.5.

4. The method of claim 1 , wherein identifying said component patch comprises identifying potential objects with a radar imaging system.

5. The method of claim 1 , wherein identifying said component patch comprises:

imaging a second ground area in front of said vehicle to produce a second ground image;

comparing said second ground image to said ground image; and

defining a component patch based on said comparison.

6. The method of claim 5 , wherein comparing said second ground image to said ground image comprises:

referencing said second ground image to said ground image;

generating an image difference by absolute image intensity difference; and

focusing upon an area of said image difference displaying a change in image intensity.

7. The method of claim 5 , wherein comparing said second ground image to said ground image comprises:

referencing said second ground image to said ground image;

generating an image difference comprising a vehicle motion compensated image difference; and

focusing upon an area of said image difference displaying a change in perspective characteristics.

8. A method for vehicle clear path detection using a camera comprising:

imaging a first ground area in front of the vehicle with the camera to produce a first ground image;

imaging a second ground area in front of the vehicle with the camera to produce a second ground image; and

analyzing said first and second ground images by grouping all individual objects upon ground together as part of an overall uniform limitation and subtracting the overall uniform limitation from said first and second ground images to formulate a clear path free of objects limiting travel of said vehicle including

iteratively identifying a component patch of said first ground image corresponding to an identified potential object, each iteration corresponding to a different component patch identified,

comparing said second ground image to said first ground image through vehicle motion compensated image differences,

generating, during each iteration for each of said component patches, a clear path likelihood describing a fractional confidence that said respective feature is indicative of a clear path free of all objects limiting travel of said vehicle and a detected object likelihood based on said comparison, wherein each respective patch clear path likelihood is determined by

analyzing raw features a-priori during a training stage using a training set of images to obtain distinguishing features, and

assigning the respective clear path likelihood of said respective feature based on the distinguishing features obtained during the training stage,

classifying respective ones of said patches as clear if said clear path likelihood is greater than said detected object likelihood, and

classifying respective ones of said patches as not clear if said clear path likelihood is not greater than said detected object likelihood, wherein each component patch classified as not clear comprises part of the overall uniform limitation subtracted from said first and second ground images.

9. An apparatus for vehicle clear path detection comprising:

a camera configured to generate a pixelated image; and

a control module analyzing said pixelated image by grouping all individual objects upon ground together as part of an overall uniform limitation and subtracting the overall uniform limitation from said pixelated image and generating a clear path output by

iteratively identifying a component patch of said pixelated image, each iteration corresponding to a different component patch identified,

extracting, during each iteration, a feature from respective ones of said patches, and

classifying each of said patches based upon respective ones of said features, said classifying comprising

determining, during each iteration for each of said component patches, a respective patch clear path likelihood describing a fractional confidence that said respective feature indicative of a clear path free of all objects limiting travel of said vehicle, wherein each respective patch clear path likelihood is determined by

analyzing raw features a-priori during a training stage using a training set of images to obtain distinguishing features, and

assigning the respective clear path likelihood of said respective feature based on the distinguishing features obtained during the training stage,

comparing, during each iteration for each of said component patches, said respective patch clear path likelihood to a threshold clear path confidence, said threshold clear path confidence comprising a predetermined value constant for all determined patch clear path likelihoods,

designating respective ones of said component patches as clear if respective ones of said patch clear path likelihoods are greater than said threshold clear path confidence, and

designating respective ones of said component patches as not clear if respective ones of said patch clear path likelihoods are not greater than said threshold clear path confidence, wherein each component patch designated as not clear comprises part of the overall uniform limitation subtracted from said ground image to formulate said clear path.

10. The apparatus of claim 9 , further comprising:

a radar imaging device generating data corresponding to an object in front of said vehicle; and

wherein said control module identifies said component patch on the basis of said data.

11. The apparatus of claim 9 , wherein said control module extracts said feature from said patch by applying feature recognition based on filtering said patch.

12. The apparatus of claim 9 , further comprising:

said camera configured to generate a second pixelated image; and

wherein said control module

compares said pixelated image to said second pixelated image by applying an absolute image intensity difference comparison,

identifies within said comparison a potential object as any area displaying change, and

utilizes said potential object to identify said component patch.

13. The apparatus of claim 9 , further comprising: equivalent said camera configured to generate a second pixelated image; and

wherein said control module

compares said pixelated image to said second pixelated image by applying a vehicle motion compensated image comparison,

identifies within said comparison a potential object as any area displaying change, and

utilizes said potential object to identify said component patch.

Assignments (12)

Continuity (1)