Lane feature detection in aerial images based on road geometry
An apparatus and method for lane feature detection from an image is performed according to predetermined path geometry. An image including at least one path is received. The image may be an aerial image. Map data, corresponding to the at least one path and defining the predetermined path geometry is selected. The image is modified according to the selected map data including the predetermined path geometry. A lane feature prediction model is generated or configured based on the modified image. A subsequent image is provided to the lane feature prediction model for a prediction of at least one lane feature.
1. A method for lane feature detection from an aerial image, the method comprising:
receiving an aerial image including at least one road;
generating a ground truth mask including at least one road geometry for the at least one road by accessing a width value for the at least one road from the map data, the at least one road geometry having a dimension corresponding to the width value;
modifying the aerial image according to the ground truth mask;
generating, using a processor, a lane feature prediction model based on the modified aerial image; and
providing a subsequent aerial image to the lane feature prediction model for a prediction of at least one lane feature.
2. The method of claim 1 , wherein the ground truth mask comprises mask values.
3. The method of claim 1 , wherein the mask values include a first value for at least one portion of the aerial image that correspond to the at least one road and a second value for at least one portion of the aerial image other than the at least one road.
4. The method of claim 1 , further comprising:
receiving probe data for the at least one path; and
analyzing the probe data to derive the width value.
5. The method of claim 1 , further comprising:
performing an aerial image processing algorithm on the aerial image including the at least one path, wherein the aerial image processing algorithm includes segmentation of a road surface or filtering of pixel values for a centerline for the at least one path,
wherein the width value is derived from an output of the aerial image processing algorithm.
6. The method of claim 5 , further comprising:
accessing the centerline for the at least one path from the map data.
7. The method of claim 4 , wherein the width is based on a number of lanes included in the at least one road or a functional classification of the at least one road.
8. The method of claim 1 , further comprising:
dividing the modified aerial image into a plurality of training aerial images according to a patch size, wherein each of the plurality of training aerial images includes at least a portion of the at least one path.
9. The method of claim 8 , further comprising:
rotating at least one of the plurality of training aerial images to a predetermined angle.
10. The method of claim 8 , wherein the plurality of training aerial images are spaced apart in a direction of the at least one road.
11. The method of claim 1 , further comprising:
identifying a scaling factor for a relationship between pixel size in the aerial image and geographic distance in the map data.
12. The method of claim 1 , further comprising:
receiving a navigation request; and
providing a navigation message including the prediction of at least one lane feature in response to the navigation request.
13. An apparatus for lane feature detection from an aerial image according to predetermined path geometry, the apparatus comprising:
a controller configured to identify map data corresponding to at least one road having a geometry, access a width value for the at least one road from the map data, the at least one road geometry having a dimension corresponding to the width value, and generate a ground truth mask including for the at least one road;
an aerial image editor configured to modify the aerial image according to the ground truth mask and the geometry of the at least one road; and
a lane feature model trained according to the modified aerial image and configured to identify at least one lane feature from a subsequent aerial image.
14. The apparatus of claim 13 , further comprising:
an aerial image trainer configured to identify a plurality of training patch aerial images from the modified aerial image.
15. The apparatus of claim 14 , wherein the aerial image trainer is configured rotate at least one of the plurality of training patch aerial image to a predetermined angle.
16. The apparatus of claim 13 , wherein the controller identifies a scaling factor to relate the map data and the aerial image, wherein the predetermined path geometry is sized according to the scaling factor.
17. The apparatus of claim 13 , wherein the aerial image editor modifies the aerial image according to a ground truth mask.
18. The apparatus of claim 17 , wherein the ground truth mask includes a first mask value for portions inside the predetermined path geometry and a second mask for portions outside the predetermined path geometry.
19. A non-transitory computer readable medium including instructions that when executed by a process are configured to perform:
receiving an aerial image including at least one road;
generating a ground truth mask including at least one road geometry for the at least one road and accessing a width value for the at least one road from the map data, the at least one road geometry having a dimension corresponding to the width value;
modifying the aerial image according to the ground truth mask;
generating, using a processor, a lane feature prediction model based on the modified aerial image; and
providing a subsequent aerial image to the lane feature prediction model for a prediction of at least one lane feature.