Top-down view classification in clear path detection

• US 8,718,329 B2
• Filed: 04/09/2012
• Issued: 05/06/2014
• Est. Priority Date: 04/09/2012
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations LLC (General Motors Company)
• Inventors: Zhang, Wende, Wang, Jinsong
• Primary Examiner(s): Tabatabai, Abolfazl
• Application Number:
  US13/441,979
• Publication Number:
  US 20130266186A1
• Time in Patent Office: 757 Days
• Field of Search: 382/103, 382/104, 382/236, 348/118, 348/148, 348/169, 348/837, 340/426.23, 340/901, 340/976, 340/988, 340/995.17, 340/995.25, 701/301, 701/527
• US Class Current: 382/104
• CPC Class Codes: G06V 20/588 Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road

Abstract

A method of detecting a clear path in a road of travel for a vehicle utilizing a top-down view classification technique. An input image of a scene exterior of the vehicle is captured. The captured input image represents a perspective view of the road of travel. The captured input image is analyzed. A segmented top-down image that includes potential clear path regions and potential non-clear path regions are generated. The segmented top-down image represents a viewing angle perpendicular to a ground plane. The segmented regions of the segmented top-down view are input to a classifier for identifying the clear path regions of travel. The identified clear path regions are utilized for navigating the road of travel.

Claims

1. A method of detecting a clear path of travel for a vehicle utilizing a top-down view technique, the method comprising the steps of:

capturing an input image of a scene exterior of the vehicle, the captured input image representing a perspective view of the road of travel;

analyzing the captured input image;

generating a segmented top-down image that includes potential clear path regions and potential non-clear path regions, the segmented top-down image representing a viewing angle perpendicular to a ground plane;

inputting the segmented regions of the segmented top-down view to a classifier for identifying the clear path regions of travel; and

utilizing the identified clear path regions for navigating the road of travel.

2. The method of claim 1 wherein the step of analyzing the captured input image further comprises the step of segmenting the captured input image, wherein segmenting the captured input image subdivides the captured input image according to discernable boundaries, wherein the segmented top-down image is generated from the segmented input image.

3. The method of claim 2 wherein the step of segmenting the captured input image includes performing motion analysis to captured input image, wherein the discernable boundaries are identified based on regions of the captured input image being associated with motion relative to a ground plane, wherein the regions associated with motion relative to the ground plane are identified as potential non-clear path regions.

4. The method of claim 2 wherein the step of segmenting the captured input image includes performing texture analysis to captured input image, wherein the discernable boundaries are identified based on regions of the captured input image being having texture, wherein the regions having texture are identified as potential non-clear path regions.

5. The method of claim 2 wherein the step of segmenting the captured input image includes performing color analysis to captured input image, wherein the discernable boundaries are identified based on regions of the captured input image having respective colors, wherein a respective region is identified as a potential clear path region or a potential non-clear path region based on an identified color of the respective region.

6. The method of claim 2 wherein the step of segmenting the captured input image includes performing geometric shape analysis, wherein the discernable boundaries are identified based on regions of the captured input image having respective geometric shapes, wherein a respective region is identified as a potential clear path region or a potential non-clear path region based on an identified geometric shape of the respective region.

7. The method of claim 2 wherein the step of segmenting the captured input image includes performing one of at least motion analysis, texture analysis, color analysis, and geometric shape analysis for subdividing the capture input image according to discernable boundaries, wherein results of identifying the discernable boundaries are input to a fusion module for generating a fused set of discernable boundaries.

8. The method of claim 7 wherein features are extracted from the segmented regions defined by the fused set of discernable boundaries, and wherein the extracted features are input to the classifier for identifying clear path regions.

9. The method of claim 2 wherein the segmented top-down image is generated based on a mapping of the pixels from the segmented input image to the segmented top-down image.

10. The method of claim 2 wherein the mapping of the pixels from the segmented input image to the segmented top-down image is generated in real-time.

11. The method of claim 1 wherein the step of analyzing the captured input image further comprises the step of generating a top-down image from the captured input image, wherein a segmented top-down image is generated from the top-down image, wherein segmenting the captured input image subdivides the captured input image according to discernable boundaries of features in the image, wherein segmenting the captured input image subdivides the capture input image according to discernable boundaries, and wherein the top-down image is a view perpendicular to a ground plane.

12. The method of claim 11 wherein the top-down image is generated based on a viewing angle perpendicular to a ground plane.

13. The method of claim 11 wherein the top-down image is generated based on a mapping of the pixels from the captured input image to the top-down image.

14. The method of claim 13 wherein the mapping of the pixels from the captured input image to the top-down image is generated in real-time.

15. The method of claim 13 wherein the mapping of the pixels from the captured input image to the top-down image is based on a pinhole camera model.

16. The method of claim 13 wherein the mapping of the pixels from the captured input image to the top-down image is based on an assumption that all pixels on are a ground plane.

17. The method of claim 1 wherein features are extracted from the segmented regions of the segmented top-down view image, and wherein the extracted features are input to the classifier for determining the clear path regions and non-clear path regions.

18. The method of claim 17 wherein a support vector machine classifies extracted features in the segmented regions of the segmented top-down view as clear path or non-clear path.

19. The method of claim 17 wherein an AdaBoost classifier classifies extracted features in the segmented regions of the segmented top-down view as clear path or non-clear path.