Lane detection systems and methods
Example lane detection systems and methods are described. In one implementation, a method received an image from-facing vehicle camera and applies a geometric transformation to the image to create a birds-eye view of the image. The method analyzes the birds-eye view of the image using a neural network, which was previously trained using side-facing vehicl camera images, to determine a lane position associated with the birds-eye view of the image.
1. A method comprising:
receiving an image from a front-facing vehicle camera;
applying, using one or more processors, a geometric transformation to the image to create a birds-eye view of the image;
analyzing, using the one or more processors, the birds-eye view of the image using a neural network to obtain a lane position associated with the birds-eye view of the image, the neural network being previously trained using side-mounted and downward-facing vehicle camera images to determine a lane position associated with the side-mounted and downward-facing vehicle camera images; and
in response to the lane position associated with the birds-eye view of the image, performing, using the one or more processors, at least one of braking, steering, accelerating, and generating a driver notification.
2. The method of claim 1 , wherein the geometric transformation transforms the received image from a first plane to a second plane.
3. The method of claim 1 , wherein the geometric transformation transforms the received image from a vertical plane to a horizontal plane.
4. The method of claim 1 , wherein the geometric transformation is an affine transformation.
5. The method of claim 1 , wherein analyzing the birds-eye view of the image includes identifying at least one lane boundary marking within the birds-eye view of the image.
6. The method of claim 5 , wherein identifying at least one lane boundary marking includes identifying at least one of a solid lane boundary, a broken lane boundary, and a dashed lane boundary.
7. The method of claim 1 , wherein analyzing the birds-eye view of the image includes separating the birds-eye view of the image into a plurality of segments.
8. The method of claim 7 , wherein analyzing the birds-eye view of the image includes analyzing each of the plurality of segments to determine a lane boundary marking within each of the plurality of segments.
9. The method of claim 8 , wherein analyzing each of the plurality of segments includes using the neural network, previously trained using side-facing vehicle camera images, to identify lane boundary marking within each of the plurality of segments.
10. The method of claim 1 , further comprising determining whether a vehicle action is required based on the lane position associated with the birds-eye view of the image.
11. The method of claim 10 , wherein the vehicle action includes communicating a warning to the driver of the vehicle.
12. The method of claim 10 , wherein the vehicle action includes activating at least one vehicle control actuator to steer the vehicle into a different lane position.
13. The method of claim 1 , wherein the method is implemented in an autonomous vehicle.
14. A method comprising:
receiving an image from a front-facing vehicle camera;
applying, using one or more processors, a geometric transformation to the image to create a birds-eye view of the image;
separating the birds-eye view of the image into a plurality of segments;
analyzing, using the one or more processors, the plurality of segments using a neural network to determine a lane position associated with the birds-eye view of the image, wherein the neural network was previously trained to determine a lane position associated with images from a side-mounted and downward-facing vehicle camera; and
in response to determining the lane position associated with the birds-eye view of the image, performing, using the one or more processors, at least one of braking, steering, accelerating, and generating a driver notification.
15. The method of claim 14 , wherein analyzing the plurality of segments includes determining a lane boundary marking within each of the plurality of segments.
16. The method of claim 15 , wherein analyzing the plurality of segments includes determining the vehicle's position within a lane of a roadway.
17. The method of claim 16 , further comprising determining whether a vehicle action is required based on the vehicle's lane position.
18. An apparatus comprising:
a front-facing camera attached to a vehicle, the front-facing camera configured to capture images representing scenes in front of the vehicle;
a geometric transformation module configured to receive an image from the front-facing camera and create a birds-eye view of the image;
an image analysis module configured to analyze the birds-eye view of the image to determine a lane position associated with the birds-eye view of the image, wherein the image analysis module uses a neural network previously trained to determine lane position using side- mounted and downward-facing vehicle camera images;
a lane position module configured to determine whether a vehicle action is required based on the lane position associated with the birds-eye view of the image; and
at least one vehicle control module confiqured to cause a vehicle control actuator to steer the vehicle into a different lane position based on the lane position associated with the birds-eye view of the image.