US 11,748,999 B2
System and method for recognizing intersection by autonomous vehicles
Soonhac Hong, San Jose, CA (US); and Jiejie Zhu, Mercer Island, WA (US)
Assigned to BEIJING JINGDONG QIANSHI TECHNOLOGY CO., LTD., Beijing (CN); and JD.COM AMERICAN TECHNOLOGIES CORPORATION, Mountain View, CA (US)
Filed by Beijing Jingdong Qianshi Technology Co., Ltd., Beijing (CN); and JD.com American Technologies Corporation, Mountain View, CA (US)
Filed on Jul. 13, 2020, as Appl. No. 16/927,422.
Prior Publication US 2022/0012507 A1, Jan. 13, 2022
Int. Cl. G06V 20/56 (2022.01); B60W 30/18 (2012.01); G06N 3/04 (2023.01); B60W 60/00 (2020.01); G06N 3/08 (2023.01); G06V 10/82 (2022.01); G01C 21/34 (2006.01); G01C 21/32 (2006.01); G01C 21/00 (2006.01)
CPC G06V 20/588 (2022.01) [B60W 30/18154 (2013.01); G06N 3/04 (2013.01); B60W 2420/42 (2013.01); B60W 2552/53 (2020.02)] 20 Claims
OG exemplary drawing
 
1. A system for autonomous navigation, comprising a visual sensor and a computing device, wherein the computing device comprises a processor and a storage device storing computer executable code, and the computer executable code, when executed at the processor, is configured to:
provide a planned path for the autonomous navigation, wherein the planned path comprises a plurality of intersections in an environment, and the intersections and roads between the intersections are represented by sequential place identifications (IDs) along the path;
receive a plurality of images of the environment along the planned path captured by the visual sensor, wherein the plurality of images comprises a current image, a previous image immediately previous to the current image, and a predetermined number of following images immediately after the current image;
perform convolutional neural network on the plurality of images to obtain a plurality of predicted place IDs of the plurality of images;
when a predicted place ID of the current image is next to a place ID of the previous image, and is the same as predicted place IDs of the predetermined number of the following images, define the predicted place ID of the current image as place IDs of the current image and the predetermined number of the following images;
when the predicted place ID of the current image is not next to the place ID of the previous image, define the place ID of the current image as the place ID of the previous image; and
perform autonomous navigation based on the planned path and the place IDs of the plurality of images,
wherein the performing convolutional neural network on the plurality of images to obtain a plurality of predicted place IDs of the plurality of images comprises: taking each of the plurality of images as an input image of the convolutional neural network, computing a probability of the input image to each place ID, and selecting a place ID with the highest probability as a predicted place ID of the input image, so as to obtain the plurality of predicted place IDs of the plurality of images.