US 11,704,913 B2
Method for automatically labeling objects in past frames based on object detection of a current frame for autonomous driving
Tae Eun Choe, Sunnyvale, CA (US); Guang Chen, Sunnyvale, CA (US); Weide Zhang, Sunnyvale, CA (US); Yuliang Guo, Sunnyvale, CA (US); and Ka Wai Tsoi, Sunnyvale, CA (US)
Assigned to BAIDU USA LLC, Sunnyvale, CA (US)
Filed by Baidu USA LLC, Sunnyvale, CA (US)
Filed on Jul. 2, 2019, as Appl. No. 16/460,192.
Prior Publication US 2021/0004643 A1, Jan. 7, 2021
Int. Cl. G06V 20/56 (2022.01); G06V 20/58 (2022.01); G06F 18/28 (2023.01); G06F 18/214 (2023.01); G06V 10/82 (2022.01)
CPC G06V 20/588 (2022.01) [G06F 18/214 (2023.01); G06F 18/28 (2023.01); G06V 10/82 (2022.01); G06V 20/58 (2022.01); G06V 20/582 (2022.01); G06V 20/584 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for labeling objects for autonomous driving, the method comprising:
receiving a list of images captured in a chronological order, wherein each of the images was captured by a sensor mounted on an autonomous driving vehicle (ADV) while driving through a driving environment;
identifying a first image of the images that includes a first object in a first dimension detected and recognized by an object detector based on the first image using an object detection algorithm; and
in response to the identified first object,
traversing the images in the list backwardly in time for a predetermined number of images within a predetermined time period to identify a second image that includes a second object in a second dimension based on a moving trail of the ADV represented by the list of images,
determining a lane configuration of a road by creating a virtual lane, wherein the lane configuration includes a number of lanes, a lane width, a lane shape and curvature, and a lane center line, wherein the determining the lane configuration of the road further comprises:
determining the number of lanes based on traffic flows of multiple streams of obstacle flows; and
determining the lane center line by tracking a moving trajectory of an obstacle, and
labeling the second object in the second image equivalent to the first object in the first image based on the lane configuration and the moving trail of the ADV automatically without user intervention, wherein the first object was detected and annotated in the first image online while the ADV was driving, wherein the second object is identified and labeled by traversing the list of images offline, wherein the list of images having the labeled second image is utilized for a subsequent object detection during autonomous driving.