| CPC B60W 60/001 (2020.02) [G06F 18/213 (2023.01); G06F 18/214 (2023.01); G06N 3/02 (2013.01); G06V 20/58 (2022.01); B60W 2420/403 (2013.01); B60W 2420/408 (2024.01)] | 16 Claims |
|
1. A method for performing long range object detection for an autonomous vehicle (AV), comprising:
using a trained neural network to process an image at a first resolution and generate a first feature map that classifies one or more objects within the image, wherein the image comprises an image captured using a sensor coupled to the AV;
cropping the image to extract a cropped section of the image;
processing, by the trained neural network, the cropped section at a second resolution that is higher than the first resolution to generate a second feature map that classifies one or more of the objects that appear within the cropped section;
cropping the first feature map to match a corresponding region of the cropped section of the image;
fusing the cropped first feature map and the second feature map to generate a third feature map,
wherein the third feature map includes object classifications from the second feature map for the one more objects that are classified in the second feature map, and object classifications from the first feature map for one or more objects that at least partially appear in the cropped first feature map but are not classified in the second feature map;
analyzing the third feature map to identify one or more objects that are positioned in a first trajectory of the AV;
outputting the object classifications for the one or more objects identified during said analyzing to a system of the AV;
performing operations by the system to generate a second trajectory for the AV based on the third feature map and object classifications; and
causing the AV to follow the second trajectory.
|