US 12,079,965 B2
Data noise reduction method and apparatus
Huaxia Xia, Beijing (CN); Zeyu Zhong, Beijing (CN); and Shenchuan Liu, Beijing (CN)
Assigned to BEIJING SANKUAI ONLINE TECHNOLOGY CO., LTD., Beijing (CN)
Filed by Beijing Sankuai Online Technology Co., Ltd., Beijing (CN)
Filed on Feb. 9, 2022, as Appl. No. 17/667,549.
Claims priority of application No. 202110336820.8 (CN), filed on Mar. 30, 2021.
Prior Publication US 2022/0318957 A1, Oct. 6, 2022
Int. Cl. G06T 5/70 (2024.01); G01S 17/89 (2020.01); G06V 10/25 (2022.01)
CPC G06T 5/70 (2024.01) [G01S 17/89 (2013.01); G06V 10/25 (2022.01); G06T 2207/10028 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A data noise reduction method, comprising:
obtaining to-be-processed point cloud data;
for each of points in the to-be-processed point cloud data, determining a point feature corresponding to the point, the point feature comprising at least one of a spatial distribution difference feature between the point and other points adjacent to the point or a point group distribution feature corresponding to a point group comprising all points in a local space in which the point is located;
recognizing a noise point from the to-be-processed point cloud data according to point features corresponding to the points in the to-be-processed point cloud data; and
performing noise reduction on the to-be-processed point cloud data according to the recognized noise point; and, wherein performing noise reduction on the to-be-processed point cloud data according to the recognized noise point comprises:
removing the recognized noise point from the to-be-processed point cloud data to obtain target point cloud data;
inputting the target point cloud data into a predetermined target object recognition model to recognize point cloud data corresponding to one or more target objects from the target point cloud data; and
removing other point cloud data than the recognized point cloud data corresponding to the one or more target objects from the target point cloud data to obtain to-be-processed point cloud data after noise reduction.