万引き犯の行動分析と検知に関する研究 Study of shoplifters’ behavior analysis tudy and shoplifting detection

博士論文

2008

Although the number of crimes in Japan has decreased since 2003, it is over2,000,000 per year now. In the U.S. and Europe, theoretical studies mainly fromthe aspect of crime-prevention have been proceeding. For example, the crimeprevention through environmental design (CPTED)theory, the crime opportunitytheory, the broken windows theory and other theories were published.On the other hand, security companies have promoted research anddevelopment of alarm systems including various sensors in order to detect anintruder into an office or home. However crime cannot be absolutely avoided inadvance even now.In this paper, shoplifters’ behavior analysis and shoplifting detection are studied.First, reports published by public institutions are surveyed. Second,crime-prevention theories are studied. Third, retired policepersons andplain-clothes security agents against shoplifting were interviewed. Fourth, theexisting measures against shoplifting are surveyed, and their effects andproblems are showed.The existing measures against shoplifting face many problems, and a drasticsolution to shoplifting is required seriously. Therefore, the feasibility of thesystem which effectively identifies shoplifters by itself is studied in this paper.Specifically, the method for identifying shoplifters by using visitors’ behaviorhistories, sales records and camera images is formulated.The following two systems related to visitors’ behavior histories are studied inthis paper. 1) Image processing system using general-purpose cameras 2) IR sensor array systemThe system of 1) had been implemented in a real retail store for 6 months, andvisitors’ behavior histories were checked manually. As a result, the capture ratewas approximately 90 %. Using this system, residence time when purchasing orshoplifting has been measured.In case of 2), behavioral records are measured by the voltage data detected fromthe IR sensor array. The experiment shows that sensors should be fixed on aceiling by bus arrangement. Specifically, up to about 10 sensors can beconnected in a bus line, and the interval between sensors should beapproximately 50 cm.The technique to analyze all recorded image has been applied to the seriouscrime such as homicide. A result of research shows that the method formulatedin this paper is well practicable. If the method formulated here is used, it is mucheasier to check security camera images every day, and undiscovered shopliftingmay be founded. The time for checking security camera images is between 17minutes and 33 minutes per day on the assumption of the following conditions.1) Residence time when purchasing or shoplifting: 10 seconds2) Passing time in front of one commercial product: 1 second3) The number of visitors: 2,000 per dayUsing the method formulated here, employee pilferage which has been missedcan be detected by analyzing whole data recorded all through the day. If a facerecognition system is integrated to the method formulated here, a salesclerk canbe alerted that a shoplifter comes to the shop. Also, if the system detectingshoplifting by behavior pattern matching is put into practice in the future, it willbe possible to detect shoplifting without shoplifters’ criminal records.