Graffiti detection: Automated crime detection in security camera feeds

The use of CCTV is prevalent in many towns and cities around the developed world. Used for their deterrent, real-time intelligence and evidence providing abilities, they serve a useful purpose, especially in social and open settings. However, for specific problem areas, or local hotspots of anti-social behaviour, a more focused solution is desirable: specific camera positions; set angles of view; picture clarity; and recording ability.
The authors approach the detection of graffiti from two directions: permanent pixel colour change in successive video frames; and assessment of the involvement of humans with an Omega (Ω) head detection process.
The pilot study, using MathLab, utilises principles of 3D volumetric video analysis, boundary and edge detection, Omega shape detection and mean shift colour contour detection to produce a successful working model for recorded video streams. The next stage of this work is to transfer the processes to on-board processors to analyse live camera feeds.