Model-based incident detection system with motion classification
AssigneeSouthwest Research Institute
InventorsMichael P. Rigney, Michael J. Magee, Ernest A. Franke
Life of this patent
Estimating the term…
Abstract
Surveillance apparatus and methods effective for detecting incidents are based upon improved image processing techniques applied to infrared and visible light spectrum images in a time sequence. A reference image is generated by removing motion objects from an image of a region of interest. The reference image is compared to an image to be analyzed to detect motion. Detected motion is analyzed and classified as motion objects. The motion objects are then compared to a motion model, which classifies the objects as being anticipated or unanticipated.
Claims16
A method of using video images to monitor incidents in a region of interest, comprising the steps of:
- computing a reference image from a set of images by removing moving objects from the set of images;
- storing a motion model representing anticipated motion in the region of interest;
- acquiring a image to be analyzed;
- computing a temporal difference image by comparing the image to be analyzed with the reference image;
- repeating the step of computing a temporal difference image to obtain a set of temporal difference images;
- calculating, from the set of temporal difference images, at least one temporal difference statistic for each pixel;
- detecting motion in a temporal difference image by separating motion pixels from background in the temporal difference image, using the temporal difference statistic;
- grouping motion pixels into motion objects;
- extracting features from the objects;
- comparing the features of motion objects to the motion model;
- wherein the motion model classifies objects as anticipated or not anticipated;
- and
- wherein the motion model further classifies anticipated objects as being of interest or not of interest.
Dependencies are inferred from the claim text.