Piece-wise network structure for long range environment perception

1. A method for environment perception, the method comprising:

receiving a point cloud from a sensor, the point cloud including a plurality of points representing positions of objects relative to the sensor;

dividing the point cloud into a plurality of distance ranges;

processing the point cloud at a first distance range of the plurality of distance ranges with a first section of a plurality of sections of a piece-wise network structure;

processing the point cloud at a second distance range of the plurality of distance ranges with a second section of the plurality of sections of the piece-wise network structure, wherein the second distance range is farther from the sensor than the first distance range; and

outputting environment perception data from the piece-wise network structure, including outputting fine detail environment perception data from the first section of the plurality of sections of the piece-wise network structure, and outputting coarse detail environment perception data from the second section of the plurality of sections of the piece-wise network structure.

2. The method of claim 1 , wherein the environment perception data includes object detection and classification, wherein the fine detail environment perception data includes object detection and classification into a first plurality of classes, wherein the coarse detail environment perception data includes object detection and classification into a second plurality of classes, and wherein the second plurality of classes is smaller than the first plurality of classes.

3. The method of claim 1 , wherein the environment perception data includes object detection and classification, wherein the fine detail environment perception data includes object detection and classification into a plurality of classes, and wherein the coarse detail environment perception data includes object detection only.

4. The method of claim 1 , wherein the environment perception data includes one or more of object detection, object classification, object tracking, free space segmentation, drivable space segmentation, or pose estimation.

5. The method of claim 1 , wherein the sensor is a LiDAR sensor, and wherein the piece-wise network structure is a piece-wise convolutional deep neural network.

6. The method of claim 5 , wherein each section of the plurality of sections of the piece-wise network structure has a different structure.

7. The method of claim 5 , wherein each section of the plurality of sections of the piece-wise network structure shares one or more middle layers of a piece-wise convolutional deep neural network.

8. The method of claim 1 , wherein dividing the point cloud into a plurality of distance ranges comprises:

dividing the point cloud into a plurality of distance ranges based on a density of point cloud data in the point cloud.

9. An apparatus configured to perform environment perception, the apparatus comprising:

a memory configured to receive a point cloud from a sensor; and

one or more processors implemented in circuitry, the one or more processors in communication with the memory and configured to:

receive the point cloud from the sensor, the point cloud including a plurality of points representing positions of objects relative to the sensor;

divide the point cloud into a plurality of distance ranges;

process the point cloud at a first distance range of the plurality of distance ranges with a first section of a plurality of sections of a piece-wise network structure;

process the point cloud at a second distance range of the plurality of distance ranges with a second section of the plurality of sections of the piece-wise network structure, wherein the second distance range is farther from the sensor than the first distance range; and

output environment perception data from the piece-wise network structure, including outputting fine detail environment perception data from the first section of the plurality of sections of the piece-wise network structure, and outputting coarse detail environment perception data from the second section of the plurality of sections of the piece-wise network structure.

10. The apparatus of claim 9 , wherein the environment perception data includes object detection and classification, wherein the fine detail environment perception data includes object detection and classification into a first plurality of classes, wherein the coarse detail environment perception data includes object detection and classification into a second plurality of classes, and wherein the second plurality of classes is smaller than the first plurality of classes.

11. The apparatus of claim 9 , wherein the environment perception data includes object detection and classification, wherein the fine detail environment perception data includes object detection and classification into a plurality of classes, and wherein the coarse detail environment perception data includes object detection only.

12. The apparatus of claim 9 , wherein the environment perception data includes one or more of object detection, object classification, object tracking, free space segmentation, drivable space segmentation, or pose estimation.

13. The apparatus of claim 9 , wherein the sensor is a LiDAR sensor, and wherein the piece-wise network structure is a piece-wise convolutional deep neural network.

14. The apparatus of claim 13 , wherein each section of the plurality of sections of the piece-wise network structure has a different structure.

15. The apparatus of claim 13 , wherein each section of the plurality of sections of the piece-wise network structure shares one or more middle layers of a piece-wise convolutional deep neural network.

16. The apparatus of claim 9 , wherein to divide the point cloud into a plurality of distance ranges, the one or more processors are further configured to:

divide the point cloud into a plurality of distance ranges based on a density of point cloud data in the point cloud.

17. The apparatus of claim 9 , wherein the apparatus comprises an automobile that includes the sensor.

18. An apparatus configured to perform environment perception, the apparatus comprising:

means for receiving a point cloud from a sensor, the point cloud including a plurality of points representing positions of objects relative to the sensor;

means for dividing the point cloud into a plurality of distance ranges;

means for processing the point cloud at a first distance range of the plurality of distance ranges with a first section of a plurality of sections of a piece-wise network structure;

means for processing the point cloud at a second distance range of the plurality of distance ranges with a second section of the plurality of sections of the piece-wise network structure, wherein the second distance range is farther from the sensor than the first distance range; and

means for outputting environment perception data from the piece-wise network structure, including means for outputting fine detail environment perception data from the first section of the plurality of sections of the piece-wise network structure, and means for outputting coarse detail environment perception data from the second section of the plurality of sections of the piece-wise network structure.

19. The apparatus of claim 18 , wherein the environment perception data includes object detection and classification, wherein the fine detail environment perception data includes object detection and classification into a first plurality of classes, wherein the coarse detail environment perception data includes object detection and classification into a second plurality of classes, and wherein the second plurality of classes is smaller than the first plurality of classes.

20. The apparatus of claim 18 , wherein the means for dividing the point cloud into a plurality of distance ranges comprises:

means for dividing the point cloud into a plurality of distance ranges based on a density of point cloud data in the point cloud.