Road-edge detection

• US 8,099,213 B2
• Filed: 07/18/2008
• Issued: 01/17/2012
• Est. Priority Date: 07/18/2008
• 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, Sadekar, Varsha
• Primary Examiner(s): Dickson, Paul N.
• Assistant Examiner(s): Spisich, George D.
• Application Number:
  US12/175,634
• Publication Number:
  US 20100017060A1
• Time in Patent Office: 1,278 Days
• Field of Search: 701/41, 701/301, 382/103, 382/104, 382/106, 382/107, 382/108, 340/435, 340/436, 340/437, 356/4.01, 356/4.08
• US Class Current: 701/41
• CPC Class Codes: G01S 17/89 for mapping or imaging
• CPC Class Codes: G01S 17/931 of land vehicles
• CPC Class Codes: G01S 7/4802 using analysis of echo signal for target characterisation; Target signature; Target cross-section
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 2207/30256 Lane; Road marking
• CPC Class Codes: G06T 7/73 using feature-based methods

Abstract

A method is provided for detecting road-side edges in a road segment using a light-based sensing system. Input range data is captured using the light-based sensing system. An elevation-based road segment is generated based on the captured input range data. The elevation-based road segment is processed by filtering techniques to identify the road segment candidate region and by pattern recognition techniques to determine whether the candidate region is a road segment. The input range data is also projected onto the ground plane for further validation. The line representation of the projected points are identified. The line representation of the candidate regions is compared to a simple road/road-edge model in the top-down view to determine whether the candidate region is a road segment with its edges. The proposed method provides fast processing speeds and reliable detection performance for both road and road-side edges simultaneously in the captured range data.

Claims

1. A method of detecting road-side edges in a road segment using a light-based sensing system, the method comprising the steps of:

capturing input range data using the light-based sensing system;

generating an elevation-based road segment based on the captured input range data;

convolving the elevation-based road segment with a differential filter for generating a filter response for identifying a road segment candidate region, the road segment candidate region including a plurality of data points that form a candidate road segment, a first road-side edge candidate, and a second road-side edge candidate;

calculating a standard deviation of the plurality of data points forming the road segment candidate region;

determining an objective value for the road segment candidate region as a function of the standard deviation of the road segment candidate region and a number of data points within road segment candidate region; and

comparing the objective value to a threshold for determining whether the road segment candidate region is a road segment.

2. The method of claim 1 further comprising the step of applying a false alarm mitigation analysis for verifying whether the road segment candidate region is the road segment, the false alarm mitigation analysis further comprises the steps of:

determining a distance between the first road-side edge candidate and the second road-side edge candidate;

comparing the distance to a predetermined distance; and

determining that the road segment candidate region is the road segment in response to the distance being greater than the predetermined distance, otherwise determining that the road segment candidate region is not the road segment.

3. The method of claim 1 wherein the first road-side edge candidate is identified by a negative peak in the filter response and the second road-side edge candidate is identified by a positive peak in the filter response.

4. The method of claim 3 wherein the filtered response includes a plurality of substantially horizontal data points and a plurality downward shifting data points, and wherein the negative peak is identified at an abrupt shifting downward of data points in the filtered response in comparison to substantially horizontal data points juxtaposed to the downward shifting data points.

5. The method of claim 4 wherein the filtered response includes a plurality of substantially horizontal data points and a plurality of upward shifting data points, and wherein the positive peak is identified at an abrupt shifting upward of data points in the filtered response in comparison to the substantially horizontal data points juxtaposed to the upward shifting data points.

6. The method of claim 1 wherein the step of determining a standard deviation includes calculating a weighted standard deviation of the road segment candidate region.

7. The method of claim 1 further comprising the step of performing a road-side edge validation routine for verifying that the candidate road-side edges are road-side edges, the road-side edge validation routine further comprising the steps of:

projecting the input range data onto a ground plane;

determining whether the first road-side edge candidate and the second road-side edge candidate are substantially perpendicular to the candidate road segment; and

validating the first road-side candidate edge and the second candidate road-side edge as road-side edges in response to the first road-side candidate edge and the second candidate road-side edge being substantially perpendicular to the candidate road segment.

8. The method of claim 1 wherein determining the objective value (f) is derived from the following formula:

f=α

*σ

_Z+γ

/N

where σ

_Z is the weighted standard deviation of the road segment candidate region, α

is a balancing parameter applied to the weighted standard deviation of the road segment candidate region, N is the number of data points within the road segment candidate region, and γ

is a balancing parameter applied to the number of data points (N) within the road segment candidate region.

9. The method of claim 8 wherein the respective balancing parameters are determined by a classifier.

10. The method of claim 9 wherein the respective balancing parameters are determined by a support vector machine classifier.

11. The method of claim 9 wherein the respective balancing parameters are determined by a neural network-base training program.

12. The method of claim 8 wherein the respective balancing parameters are selected to provide a balance between the weighted standard deviation and the number of data points for a road segment candidate region.

13. The method of claim 1 wherein the road segment is provided to an autonomous steering system for providing locations of the road-side edges for maintaining road position.

14. The method of claim 1 wherein the road segment is provided to an image display device for providing the location of the road-side edges in an image display device.

15. A road-side edge detection system comprising:

a light-based sensing device for capturing range input data;

a processor for receiving the captured range input data, the processor generating elevation-based road segment data in response to the captured road input data, the processor convolving the elevation-based road segment data with a filter for identifying a road segment candidate region that contains a plurality of data points that form a candidate road segment, a first edge candidate segment, and a second edge candidate segment, the processor determining a weighted standard deviation of the plurality of data points forming the elevation-based road segment candidate region, the processor calculating an objective value for the road segment candidate region as a function of the weighted standard deviation and a number of data points within the road segment candidate region, the processor comparing the objective value to a threshold for determining whether a road segment candidate region is a road segment; and

an output device identifying a location of each of the road-side edges of the road segment.

16. The road-side edge detection system of claim 15 further comprising a false alarm mitigation module for verifying whether a road segment candidate region is a road segment wherein a distance determined between the first edge candidate region and the second edge candidate region is compared to a predetermined distance, and wherein the segment is determined to be the road segment if the distance is greater than the predetermined distance, otherwise determining that the region is not the road segment.

17. The road-side edge detection system of claim 15 further comprising a classifier for generating the weighting parameters.

18. The road-side edge detection system of claim 17 wherein the classifier is a support vector machine classifier.

19. The road-side edge detection system of claim 17 wherein the classifier is a neural network training program.

20. The road-side edge detection system of claim 15 wherein the light-based sensing device includes a LIDAR sensing device.