Automatic detection of suspicious objects using surveillance cameras
Masayuki Daikoku, Stephen Karungaru, Kenji Terada · Society of Instrument and Control Engineers of Japan · 2013
Suspicious objects must be detected at important sensitive institutions like airports, train stations, sports arenas, etc. to maintain safety at these locations. In this paper, we propose a method for automatic detection of suspicious object using a surveillance camera. A person and object areas are extracted using the current frame, previous frame and background. A suspicious object is detected using Histograms of Oriented Gradients (HOG) feature detection method. These features are then learned using the AdaBoost algorithm. Using data collected in our laboratory, the system achieved an average accuracy of 85% for 3 types of objects while operating in real time.