Optimization of human activity determination from video

A computer system for automated analysis of video data for determination of human behavior, comprising:

• a central processing unit, computer readable memory and a computer readable storage media;
• first program instructions to segment a video stream of images of a cashier in a retail scanning lane into a plurality of discrete individual frame image primitives comprising:
• pick-up primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item pick-up region of interest of the retail scanning lane video stream images;
• scan primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item scanning region of interest of the retail scanning lane video stream images that is different from the pick-up region of interest;
• and
• drop primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item drop region of interest of the retail scanning lane video stream images that is different from the pick-up and scanning regions of interest;
• second program instructions to combine a set of one each of the pick-up, scan and drop primitives that are sequential and closest in time to each other into a visual event, wherein the visual event pick-up primitive is prior in time to the visual event scan primitive, and the visual event scan primitive is prior in time to the visual event drop primitive;
• and
• third program instructions to determine a normalized visual quality value of one or zero for an average of feature vectors over an adjacent pair of the visual event primitives as a function of a Gaussian mixture model, and to set a binary fake scan indicator variable for the visual event to true if the determined normalized visual quality value is one, or to false if the determined normalized visual quality value is zero;
• and
• fourth program instructions to process the visual event as a function of the binary fake scan indicator variable and of non-video retail transaction log data associated with the video stream that lists items scanned in the scanning lane, by:
• associating the visual event with one of a plurality of logged item bar code scan transactions of the non-video transaction data if the visual event is aligned with the one logged transaction within a Gaussian transformed temporal distance constraint;
• issuing an alert that the visual event is a fake scan if the binary fake scan indicator variable is true and the visual event does not align with any of the logged transactions within the Gaussian transformed temporal distance constraint;
• and
• dropping the visual event if the binary fake scan indicator variable is false and the visual event does not align with any of the logged transactions within the Gaussian transformed temporal distance constraint;
• wherein the first, second, third and fourth program instructions are stored on the computer readable storage media for execution by the central processing unit via the computer readable memory.

A computer program product for automated analysis of video data for determination of human behavior, said computer program product comprising:

• a computer readable tangible storage medium which is not a signal and has computer readable program code embodied therewith, the computer readable program code comprising instructions that, when executed by a computer processing unit, cause the computer processing unit to:
• segment a video stream of images of a cashier in a retail scanning lane into a plurality of discrete individual frame image primitives comprising:
• pick-up primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item pick-up region of interest of the retail scanning lane video stream images;
• scan primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item scanning region of interest of the retail scanning lane video stream images that is different from the pick-up region of interest;
• and
• drop primitive segments that each comprise sequential frames of the video stream that have motion pixels associated with movement of at least one of an item and a hand of the cashier within an item drop region of interest of the retail scanning lane video stream images that is different from the pick-up and scanning regions of interest;
• combine a set of one each of the pick-up, scan and drop primitives that are sequential and closest in time to each other into a visual event, wherein the visual event pick-up primitive is prior in time to the visual event scan primitive, and the visual event scan primitive is prior in time to the visual event drop primitive;
• and
• determine a normalized visual quality value of one or zero for an average of feature vectors over an adjacent pair of the visual event primitives as a function of a Gaussian mixture model, and to set a binary fake scan indicator variable for the visual event to true if the determined normalized visual quality value is one, or to false if the determined normalized visual quality value is zero;
• and
• process the visual event as a function of the binary fake scan indicator variable and of non-video retail transaction log data associated with the video stream that lists items scanned in the scanning lane, by:
• associating the visual event with one of a plurality of logged item bar code scan transactions of the non-video transaction data if the visual event is aligned with the one logged transaction within a Gaussian transformed temporal distance constraint;
• issuing an alert that the visual event is a fake scan if the binary fake scan indicator variable is true and the visual event does not align with any of the logged transactions within the Gaussian transformed temporal distance constraint;
• and
• dropping the visual event if the binary fake scan indicator variable is false and the visual event does not align with any of the logged transactions within the Gaussian transformed temporal distance constraint.