Oracle Corporation
Post processing to improve motion detection results

Abstract:

These techniques detect and compare amounts of motion of objects within video footage having sources of apparent motion. In an embodiment, for each frame of a video, a computer subtracts a background from the frame, converts the frame from grayscale to monochrome, and isolates regions of interest (ROIs) in the frame. For each frame, the computer identifies identifiable objects in the ROIs, selects selectable objects from the identifiable objects that have a percentage of motion that is within a particular range, and merges, into merged objects, subsets of the selectable objects that are separated by no more than a particular distance. Based on the merging, metadata is generated for any of: an object of the merged objects, the frame, or the video. The metadata may have a bounding rectangle, a bounding polygon, a frame timestamp, a frame identifier, or an object(s)'s identifier or percentage of motion.