Anomalous Behavior Detection Using the Geometrical Complex Moments in Crowd Scenes of Smart Surveillance Systems

Abstract

In this research work a data stream clustering method done by extracting regions of interest from the frames of video clips (UCSD pedestrian dataset (ped1 and ped2 datasets) video clips, and VIRAT VIDEO dataset video clips). In extraction process the HARRIS or FAST detector applied on the frames of video clips to extract list of pairs of interest points. From these pairs a list of features such as: distance, direction, x-coordinate, y-coordinate obtained to use as an input to the clustering method based on seed based region growing technique. From these clusters a regions of interest extracted according the pairs coordinates of each cluster. Finally, from these regions a set of geometrical complex moments obtained and then used in anomaly detection system. The results indicated that using HARRIS detector achieved detection rates are 7.88%, 51.30%, and 56.67% with false alarms are 19.39%, 32.61%, and 60.00% by using Ped1, Ped2, and VIRAT datasets respectively. For the case of using FAST detector, the best detection rates are 6.67%, 44.78%, 53.33% with false alarm rates are 33.33%, 41.74%, 70% by using the datasets respectively.