NCJ Number
206534
Journal
Journal of Forensic Sciences Volume: 49 Issue: 4 Dated: July 2004 Pages: 770-780
Date Published
July 2004
Length
11 pages
Annotation
In Taiwan, criminals who use an ATM (automatic teller machine) illegally will often hide their faces by wearing a safety helmet; this paper describes a safety-helmet detection technology and the testing of its application to an ATM surveillance system.
Abstract
The safety-helmet detection technology involves a circle/circular arc detection method based on the modified Hough transform (HT). The HT has been widely used for pattern detection; however, the drawback of the traditional HT is the large computation and storage requirement, making it difficult for practical application. The current paper proposes a circle/circular arc detection method (the modified HT) for the detection of safety helmets. Since the safety helmet location will be within the set of the obtainable circular arcs, if any exist, the technology uses the geometric features to determine whether any safety helmet exists in the set. This paper provides a summary description of the steps for circle detection. Centroid information is used to compute the parameters of a circular arc characteristic of a safety helmet, and geometric features are used to determine whether any safety helmet exists in the image produced. The proposed method can be used to help the surveillance systems of ATM's record customer face information more precisely; and if customers wear safety helmets to block their faces, the system can send messages to remind them to take off their helmets before the ATM will dispense money. The method can also be applied to the surveillance systems of banks by providing an early warning safeguard when any customer or intruder uses a safety helmet to avoid his/her face information from being recorded by the surveillance system. The technology and its application was satisfactorily tested, and its calibration was sufficiently precise to avoid a false positive for a man with little head hair. 13 figures and 10 references