US 12,462,559 B2
Systems and methods for semantic guided temporal activity detection and classification of videos
Orpaz Goldstein, Los Angeles, CA (US); and Sayak Nag, Riverside, CA (US)
Assigned to DRNC HOLDINGS, INC., Wilmington, DE (US)
Filed by DRNC HOLDINGS, INC., Wilmington, CA (US)
Filed on Apr. 18, 2022, as Appl. No. 17/722,658.
Prior Publication US 2023/0334857 A1, Oct. 19, 2023
Int. Cl. G06V 20/40 (2022.01); G06V 10/20 (2022.01)
CPC G06V 20/41 (2022.01) [G06V 10/255 (2022.01); G06V 20/47 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A method comprising:
determining a set of features that occur within a sequence of frames within a first set of videos, the first set of videos depicting a first activity;
determining, in response to a query for a second activity that is different from the first activity, one or more videos depicting the second activity, wherein determining the one or more videos depicting the second activity comprises analyzing a second set of videos using at least one of the set of features determined from the first set of videos depicting the first activity:
providing the one or more videos from the second set of videos determined using the at least one of the set of features determined from the first set of videos;
determining that a second video from the second set of videos comprises the at least one feature of the set of features that represent the first activity;
performing object recognition across one or more frames of the second video;
determining that the second video does not contain a particular object as a result of the object recognition; and
removing the second video from a response to the query for the one or more videos of the second activity in response to determining that the second video does not contain the particular object.