Region-based Abnormal Motion Detection in Video Surveillance
Jorge Henrique Busatto Casagrande, Marcelo Ricardo Stemmer · 2014
This article presents a method to detect abnormal motion based on the subdivision of regions of interest in the scene. The proposed model reduces the large amount of data in the motion analysis and the respective computational cost. The regions are spatially identified and contain data of n-dimensional transition vectors, resulting from the centroid tracking of multiple moving objects. On these data, we applied a one-class supervised training with only one set of normal tracks on Gaussian mixtures to find relevant clusters, which discriminate the trajectory of objects. To complete the learning model, the lowest probability of transition vectors in the test phase is used as the threshold to classify abnormal motions. The ROC (Receiver Operating Characteristic) curves are used to this task and also to determinate the efficiency of the model. In order to find the best result of our method, the process is repeated to find a new measure of efficiency for each size increment of the region grid. The results show that there is a range of grid size values, which ensure a good margin of correct abnormal motions detection for each type of scenario, even with a significant reduction of data samples.