Systems and methods for lane-marker detection

1. An electronic device, comprising:

a memory; and

a processor in communication with the memory, the processor configured to:

determine a red-green-blue (RGB)-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in an image;

extract a lane response map from the RGB-to-grayscale conversion of the image the lane response map including a plurality of lane response values, wherein a lane response value for a pixel in the lane response map is determined based on a penalty component that reduces the lane response value as a difference between neighboring pixel values increases;

transform the lane response map to an inverse-perspective-mapping (IPM) domain;

define one or more sections in the IPM domain of the lane response map;

detect one or more straight lane-marker segments in the one or more sections based on a Radon transform of the lane response map in the IPM domain; and

cluster the one or more straight lane-marker segments to obtain one or more lane markers.

2. The electronic device of claim 1 , wherein defining the multiple sections is further based on at least one of a lane-marker segment width assumption, a pre-defined detection distance range, or calibration data.

3. The electronic device of claim 1 , wherein the processor is further configured to:

perform a transform in each of the multiple sections; and

obtain a lane-marker position and a lane-marker orientation based on the transform.

4. The electronic device of claim 1 , wherein the processor is further configured to filter one or more potential straight line segments to reduce false positives.

5. The electronic device of claim 1 , wherein the processor is further configured to refine the one or more lane markers.

6. The electronic device of claim 5 , wherein the processor is further configured to optimize a score function to obtain the one or more lane markers.

7. The electronic device of claim 1 , wherein the processor is further configured to apply a fitting to the one or more lane markers.

8. The electronic device of claim 1 , wherein the processor configured to extract the lane response map from the image comprises the processor configured to perform row-wise one-dimensional (1D) filtering on the image.

9. The electronic device of claim 1 , wherein the processor configured to determine the RGB-to-grayscale conversion that maximizes intensity distances between the lane marker color and the road color in the image comprises:

convert the image from a red-green-blue (RGB) color model to a grayscale color model; and

apply a trained projection model that maximizes intensity distances between the lane marker color and the road color using a linear discriminant analysis.

10. The electronic device of claim 1 , wherein the processor is further configured to apply an ego-lane mask to the image to restrict a lane-marker search region.

11. A method, comprising:

determining a red-green-blue (RGB)-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in an image;

extracting a lane response map from the RGB-to-grayscale conversion of the image, the lane response map including a plurality of lane response values, wherein a lane response value for a pixel in the lane response map is determined based on a penalty component that reduces the lane response value as a difference between neighboring pixel values increases;

transforming the lane response map to an inverse-perspective-mapping (IPM) domain;

defining one or more sections in the IPM domain of the lane response map;

detecting one or more straight lane-marker segments in the one or more sections based on a Radon transform of the lane response map in the IPM domain; and

clustering the one or more straight lane-marker segments to obtain one or more lane markers.

12. The method of claim 11 , wherein extracting the lane response map from the image comprises performing row-wise one-dimensional (1D) filtering on the image.

13. The method of claim 11 , wherein determining the RGB-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in the image comprises:

converting the image from a red-green-blue (RGB) color model to a grayscale color model; and

applying a trained projection model that maximizes intensity distances between the lane marker color and the road color using a linear discriminant analysis.

14. The method of claim 11 , further comprising applying an ego-lane mask to the image to restrict a lane-marker search region.

15. A non-transitory computer readable medium storing computer executable code, comprising:

code for causing an electronic device to determine a red-green-blue (RGB)-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in an image;

code for causing the electronic device to extract a lane response map from the RGB-to-grayscale conversion of the image, the lane response map including a plurality of lane response values, wherein a lane response value for a pixel in the lane response map is determined based on a penalty component that reduces the lane response value as a difference between neighboring pixel values increases;

code for causing the electronic device to transform the lane response map to an inverse-perspective-mapping (IPM) domain;

code for causing the electronic device to define one or more sections in the IPM domain of the lane response map;

code for causing the electronic device to detect one or more straight lane-marker segments in the one or more sections based on a Radon transform of the lane response map in the IPM domain; and

code for causing the electronic device to cluster the one or more straight lane-marker segments to obtain one or more lane markers.

16. The computer readable medium of claim 15 , wherein the code for causing the electronic device to extract the lane response map from the image comprises performing row-wise one-dimensional (1D) filtering on the image.

17. The computer readable medium of claim 15 , wherein the code for causing the electronic device to determine the RGB-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in the image comprises:

code for causing the electronic device to convert the image from a red-green-blue (RGB) color model to a grayscale color model; and

code for causing the electronic device to apply a trained projection model that maximizes intensity distances between the lane marker color and the road color using a linear discriminant analysis.

18. The computer readable medium of claim 15 , further comprising code for causing the electronic device to apply an ego-lane mask to the image to restrict a lane-marker search region.

19. An apparatus, comprising:

means for determining a red-green-blue (RGB)-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in an image;

means for extracting a lane response map from the RGB-to-grayscale conversion of the image, the lane response map including a plurality of lane response values, wherein a lane response value for a pixel in the lane response map is determined based on a penalty component that reduces the lane response value as a difference between neighboring pixel values increases;

means for transforming the lane response map to an inverse-perspective-mapping (IPM) domain;

means for defining one or more sections in the IPM domain of the lane response map;

means for detecting one or more straight lane-marker segments in the one or more sections based on a Radon transform of the lane response map in the IPM domain; and

means for clustering the one or more straight lane-marker segments to obtain one or more lane markers.

20. The apparatus of claim 19 , wherein the means for extracting the lane response map from the image comprise means for performing row-wise one-dimensional (1D) filtering on the image.

21. The apparatus of claim 19 , wherein the means for determining the RGB-to-grayscale conversion that maximizes intensity distances between a lane marker color and a road color in the image comprise:

means for converting the image from a red-green-blue (RGB) color model to a grayscale color model; and

means for applying a trained projection model that maximizes intensity distances between the lane marker color and the road color using a linear discriminant analysis.

22. The apparatus of claim 19 , further comprising means for applying an ego-lane mask to the image to restrict a lane-marker search region.