Video analytics for retail business process monitoring

摘要

A system for video monitoring a retail business process includes a video analytics engine to process video obtained by a video camera and generate video primitives regarding the video, A user interface is used to define at least one activity of interest regarding an area being viewed, each activity of interest identifying at least one of a rule or a query regarding the area being viewed. An activity inference engine processes the generated video primitives based on each defined activity of interest to determine if an activity of interest occurred in the video.

主权项

1. A method for video monitoring a retail business process comprising:
obtaining video from a video camera; processing, by one or more processors within one or more devices, the video obtained from the video camera; generating, by one or more processors within one or more devices, video primitives regarding the video, wherein the video primitives comprise a high-value item stealing video primitive; defining at least one activity of interest regarding an area being viewed, wherein an activity of interest identifies at least one of a user-defined rule or a user-defined query regarding the area being viewed, wherein at least one of the user-defined rule or the user-defined query comprises detection of high-value item stealing based on counting, using only the video, a number of times a person removes high-value items off a shelf; wherein the number of times the person removes the high-value items off the shelf is counted using only the video by analyzing a motion pattern in the video to detect periodic motion based on an array of motion block information; wherein the array of motion block information comprises motion blocks associated with a person removing high-value items off the shelf, wherein the motion blocks are generated based on one or more foreground masks and include a predetermined number of foreground pixels, wherein the predetermined number of foreground pixels is a user-defined parameter; processing, by one or more processors within one or more devices, the generated video primitives based on at least one defined activity of interest to determine that an activity of interest occurred in the video based on determining that the number of times the person removes the high-value items off the shelf exceeds a user-defined threshold; and generating an alert based on determining that the number of times the person removes the high-value items off the shelf exceeds the user-defined threshold.