US 12,223,017 B2
Self-organized learning of three-dimensional motion data
Batuhan Okur, Singapore (SG); and Roshan Gopalakrishnan, Singapore (SG)
Assigned to Rapsodo Pte. Ltd., Singapore (SG)
Filed by RAPSODO PTE. LTD., Singapore (SG)
Filed on Aug. 27, 2021, as Appl. No. 17/460,038.
Prior Publication US 2023/0065922 A1, Mar. 2, 2023
Int. Cl. G06F 18/25 (2023.01); G01S 13/08 (2006.01); G01S 13/58 (2006.01); G01S 13/86 (2006.01); G01S 13/89 (2006.01); G06T 7/215 (2017.01); G06T 7/80 (2017.01); G06V 40/20 (2022.01)
CPC G06F 18/251 (2023.01) [G01S 13/08 (2013.01); G01S 13/587 (2013.01); G01S 13/867 (2013.01); G01S 13/89 (2013.01); G06T 7/215 (2017.01); G06T 7/80 (2017.01); G06V 40/23 (2022.01); G06T 2207/10028 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30224 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
training, with at least one processor, a machine learning model in a training environment, to:
identify and track a moving object based on training image data, and
identify movement signature of the moving object based on training radar data;
capturing image data associated with the moving object in an analyzed environment at a plurality of points in time;
capturing radar data associated with the moving object in the analyzed environment at the plurality of points in time;
obtaining, by the trained machine learning model, the image data and the radar data associated with the moving object in the analyzed environment;
pairing, with the at least one processor, each image datum with a corresponding radar datum based on a chronological occurrence of the image data and the radar data in the analyzed environment; and
generating, by the trained machine learning model, a three-dimensional motion representation relating to the moving object that is associated with the image data and the radar data;
wherein a number of physical devices configured to collect training data in the training environment is greater than a number of physical devices in the analyzed environment and wherein the physical devices include at least one camera unit and one radar unit;
wherein the trained machine learning model is configured to recognize and distinguish between two or more different moving objects based on sensor data obtained from the analyzed environment;
wherein the trained machine learning model is configured to match the image data of each of the two or more different moving objects with the radar data captured at same points in time with the image data, wherein the trained machine learning model is configured to determine that the radar data corresponding to first one of the two or more different moving objects includes first frequency signature characteristics indicating movement corresponding to the first one of the two or more different moving objects, and that the radar data corresponding to second one of the two or more different moving objects excludes the first frequency signature characteristics.