Abnormal event detection based on IPZM
Biao Yin · 2011
This paper presents a new method to detect events in the Self-service Banking. The method reduces two-person interactions to event semantics template match. Firstly, the method gets the objects using the back-ground subtraction algorithm and counts the person in the video. Then, the computation could be reduced by the symmetry of Fourier kernel function of the improved pseudo-Zernike moment. The event semantics template and a shape description vector consists of seven IPZM are combined to detect events at last. Through the experiment, this method is proved to be effective with the three indicators of precision A, recall R and frame-rate F on detecting three kinds of events, such as normal event, standing one by one when withdrawing and violent robbery.