Automatic Visual Recognition of Armed Robbery particularly in convenience store settings
Specific Target______________________ one arm “aiming” at another person, perpendicular to body both arms raised “defensively” or in a “hold-up” position, parallel to and above body classic hold-up poses
Procedure_________________________ 1.) find people -- isolate blobs 2.) segment resultant blobs 3.) signal trouble situations
1.) Isolate Blobs_____________________ a.) find candidate pixels -- background subtraction b.) group related pixels -- connected components c.) refine output -- dilation/erosion, error checking, etc. background subtraction and final output on frame 240 of 460
2.) Segment Blobs___________________ a.) skeleton calculation b.) isolate separate curves c.) identify different body parts
2.) Segment Blobs - continued__________ segmentation of skeleton; individual segments numbered (colored) sequentially; color of POI indicates number of segments attached segment identification; 2 arms, 1 head, 1 torso, 2 legs identified; linked legs successfully separated
3.) Signal Trouble Situations___________ a.) identify arm positions b.) confirm history and other factors indicate potential threat (not yet complete) c.) generate appropriate output -- alert monitor, call police, release chloroform, etc.