Detection methods to detect objects such as bicyclists
To reliably detect an object such as a bicycle at increased range, an Advanced Driving Support System uses a deep neural network(s) to process an ambient (grey-scale) image into an object that is then tracked by a second range detection camera. Most objects of interest, such as bicycles and automobiles, are outfitted with one or more retroreflectors that are used to cue the neural network to the object of most interest. As the retroreflectors also tend to saturate the range detection camera, a method is used to manage the saturation and estimate the correct range to the object.
1. A method for detecting an object comprising:
generating, using at least one sensor, a first amplitude map including a first detection region indicating a retroreflector,
generating, using the at least one sensor, a raw point cloud including a second detection region indicating the retroreflector,
using at least one processor to perform operations comprising:
generating a merged point cloud based on the raw point cloud and the first amplitude map, the merged point cloud including corrected distance values in the second detection region, and
processing, using a deep learning model, the first amplitude map to classify an object of interest that includes the first detection region.
2. The method of claim 1 , wherein the merged point cloud is generated by merging, using the at least one processor-, the first amplitude map a corrected point cloud including the corrected distance values.
3. The method of claim 2 , wherein generating of the corrected point cloud includes inferring the corrected distance values based on distance values of pixels neighboring the second detection region.
4. The method of claim 2 , wherein
the first detection region includes saturated pixels,
the second detection region includes corresponding pixels that correspond to the saturated pixels, and
generating of the corrected point cloud includes replacing distance values of the corresponding pixels with the corrected distance values.
5. The method of claim 4 , wherein the corrected distance values are mean distance values of pixels that border the first detection region.
6. The method of claim 1 , wherein the deep learning model divides the first amplitude map into a grid wherein each cell in the grid detects objects within itself.
7. The method of claim 1 , wherein the deep learning model separates different objects in the first amplitude map and returns pixels of the first amplitude map that each object occupies.
8. The method of claim 1 , wherein the operations further comprise detecting an approaching object based on i) the second detection region and ii) a classification of the object of interest.
9. The method of claim 1 , wherein the first amplitude map comprises an image detected by a camera.
10. The method of claim 1 , wherein the operations further comprise generating the merged point cloud with a cue of the retroreflector.
11. The method of claim 1 , wherein the operations further comprise generating a second amplitude map, the generating of the second amplitude map including modifying amplitude values of the first amplitude map.
12. The method of claim 1 , wherein
the object of interest is classified as a bicycle, and
the operations further comprise detecting the bicycle from a range of at least 20 meters based on the merged point cloud.
13. The method of claim 1 , wherein the operations further comprise:
detecting a retroreflective surface of the retroreflector in the first amplitude map as a cue of the object of interest,
removing deleterious saturation effects caused by the retroreflective surface in the first amplitude map to generate a second amplitude map including the retroreflective surface,
merging the second amplitude map containing the object of interest with a corrected point cloud to generate the merged point cloud, and
determining at least one point of the merged point cloud that corresponds to the object of interest.
14. A system for detecting an object at increased range comprising:
at least one sensor configured to generate
a first amplitude map including a first detection region indicating a retroreflector, and
a raw point cloud including a second detection region indicating the retroreflector; and
at least one processor configured to perform operations comprising:
generating a merged point cloud based on the raw point cloud and the first amplitude map, the merged point cloud including corrected distance values in the second detection region, and
processing, using a deep learning model, the first amplitude map to classify an object of interest that includes the first detection region.
15. The system of claim 14 , wherein the merged point cloud is generated by merging the first amplitude map a corrected point cloud.
16. The system of claim 15 , wherein generating of the corrected point cloud includes inferring the corrected distance values based on distance values of pixels neighboring the second detection region.
17. The system of claim 15 , wherein
the first detection region includes saturated pixels,
the second detection region includes corresponding pixels that correspond to the saturated pixels, and
generating of the corrected point cloud includes replacing distance values of the corresponding pixels with the corrected distance values.
18. The system of claim 17 , wherein the corrected distance values are mean distance values of pixels that border the first detection region.
19. The system of claim 14 , wherein the deep learning model is configured to divide the first amplitude map into a grid wherein each cell in the grid detects objects within itself.
20. The system of claim 14 , wherein the deep learning model is configured to separate different objects in the first amplitude map and return pixels of the first amplitude map that each object occupies.
21. The system of claim 14 , wherein the operations further comprise detecting an approaching object based on i.) the second detection region and ii.) a classification of the object of interest.
22. The system of claim 14 , wherein the first amplitude map comprises an image detected by a camera.
23. The system of claim 14 , wherein the operations further comprise generating the merged point cloud with a cue of the retroreflector.
24. The system of claim 14 , wherein the operations further comprise generating a second amplitude map, the generating of the second amplitude map including modifying amplitude values of the first amplitude map.
25. The system of claim 14 , wherein
the object of interest is classified as a bicycle, and
the operations further comprise detecting the bicycle from a range of at least 20 meters based on the merged point cloud.
26. The system of claim 14 , wherein the operations further comprise:
detecting a retroreflective surface of the retroreflector in the first amplitude map as a cue of the object of interest,
removing deleterious saturation effects caused by the retroreflective surface in the first amplitude map to generate a second amplitude map including the retroreflective surface,
merging the second amplitude map containing the object of interest with the a corrected point cloud to generate the second point cloud, and
determining at least one point of the of the merged point cloud that corresponds to the object of interest.