US 12,272,073 B2
System and method for real-time object detection in video
Eui-Nam Huh, Yongin-si (KR); Md. Alamgir Hossain, Yongin-si (KR); Dao Tan Tri Nguyen, Yongin-si (KR); Md. Delowar Hossain, Yongin-si (KR); and Ga-Won Lee, Suwon-si (KR)
Assigned to UNIVERSITY-INDUSTRY COOPERATION GROUP OF KYUNG HEE UNIVERSITY, Yongin-si (KR)
Filed by UNIVERSITY-INDUSTRY COOPERATION GROUP OF KYUNG HEE UNIVERSITY, Yongin-si (KR)
Filed on Feb. 8, 2022, as Appl. No. 17/666,843.
Claims priority of application No. 10-2021-0039957 (KR), filed on Mar. 26, 2021.
Prior Publication US 2022/0309679 A1, Sep. 29, 2022
Int. Cl. G06T 7/194 (2017.01); G06T 5/70 (2024.01); G06T 7/11 (2017.01); G06T 7/90 (2017.01)
CPC G06T 7/194 (2017.01) [G06T 5/70 (2024.01); G06T 7/11 (2017.01); G06T 7/90 (2017.01)] 10 Claims
OG exemplary drawing
 
1. A method for extracting a moving object in a video by an object detection system operated by at least one processor, the method comprising:
receiving a certain unsmoothed frame included in the video;
when the certain frame is a first frame of the video, initializing a background sample for a target pixel in the first frame, and storing the initialized background sample as a reference background sample, the reference background sample including a plurality of neighboring pixels of the target pixel as a plurality of reference pixels, and the neighboring pixels being pixels within a window having a predetermined size;
when the certain frame is not the first frame of the video, calculating a pixel intensity difference between each of the plurality of reference pixels of the reference background sample extracted from the first frame of the video and the target pixel of the certain frame;
counting a number of times the pixel intensity difference is less than or equal to a predetermined threshold in the plurality of reference pixels, and determining the number of reference pixels as a similarity measurement value; and
segmenting the target pixel into the background or the foreground according to the similarity measurement value, and extracting the foreground from the certain frame as an object.