US 12,450,922 B2
Method for detecting an interior condition of a vehicle, and system implementing the same
Yi-Feng Su, Lugong (TW); Zi-Rong Ding, Lugong (TW); and Meng-Jiang He, Lugong (TW)
Assigned to AUTOMOTIVE RESEARCH & TESTING CENTER, Lugong (TW)
Filed by Automotive Research & Testing Center, Lugong (TW)
Filed on Dec. 15, 2022, as Appl. No. 18/082,126.
Claims priority of application No. 111136796 (TW), filed on Sep. 28, 2022.
Prior Publication US 2024/0104943 A1, Mar. 28, 2024
Int. Cl. G06V 20/59 (2022.01); G01S 13/89 (2006.01); G06T 5/70 (2024.01); G06T 5/73 (2024.01); G06T 7/70 (2017.01); G06V 40/16 (2022.01); G08B 21/00 (2006.01)
CPC G06V 20/59 (2022.01) [G01S 13/89 (2013.01); G06T 5/70 (2024.01); G06T 5/73 (2024.01); G06T 7/70 (2017.01); G06V 40/171 (2022.01); G08B 21/00 (2013.01); G06T 2207/20044 (2013.01); G06T 2207/30268 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method for detecting an interior condition of a vehicle, comprising steps of:
A) by a computing device, receiving, from a radar device located in an interior space of the vehicle and electrically connected to the computing device, a plurality of candidate point cloud datasets that are acquired within a time period by the radar device with respect to the interior space of the vehicle, and acquiring a target point cloud dataset based on the candidate point cloud datasets;
B) by the computing device, receiving, from a camera device located in the interior space of the vehicle and electrically connected to the computing device, a plurality of candidate images of the interior space of the vehicle that are captured within the time period by the camera device, and acquiring a target image from among the candidate images;
C) by the computing device, acquiring a synthesized image based on the target point cloud dataset and the target image;
D) by the computing device, using a living-object detection model to obtain a living-object position dataset, which is related to a position of a living object in the synthesized image;
E) by the computing device, using a skeleton detection model to obtain a skeleton feature dataset based on the living-object position dataset, wherein the skeleton feature dataset is related to a position of a skeleton of the living object in the synthesized image;
F) by the computing device, using a face detection model to obtain a facial feature dataset based on the synthesized image and the living-object position dataset, wherein the facial feature dataset is related to positions of multiple facial features of the living object in the synthesized image; and
G) by the computing device, determining the interior condition of the vehicle based on the facial feature dataset and the skeleton feature dataset,
wherein step A) includes sub-steps of:
A-1) receiving a point cloud dataset from the radar device at a current time point within the time period, and making the point cloud dataset thus received serve as one of the candidate point cloud datasets;
A-2) adding some of the candidate point cloud datasets that have been received since a starting point of the time period together, and removing outliers from said some of the candidate point cloud datasets that have been added together, so as to obtain a synthesized point cloud dataset;
A-3) determining whether a total number of said some of the candidate point cloud datasets is greater than a predetermined number;
A-4) upon determining that the total number of said some of the candidate point cloud datasets is not greater than the predetermined number, repeating sub-steps A-1) to A-3); and
A-5) upon determining that the total number of said some of the candidate point cloud datasets is greater than the predetermined number, making the synthesized point cloud dataset serve as the target point cloud dataset.