Abnormal event detection based on SVM in video surveillance

Abstract

In order to increase the accuracy of abnormal event detection in crowd video surveillance, this paper proposes a novel hybrid optimization of feature selection and support vector machine (SVM) training model based on genetic algorithm. For reducing dimensions of multi-feature, we propose an adaptive genetic simulated annealing algorithm (ASAGA) feature selection method. The ASAGA takes advantage of the local search ability of simulated annealing algorithm (SA) to solve the slow convergence and high complexity drawbacks of genetic algorithm (GA). And also improve the SVM training model performance based on genetic algorithm. Experimental results demonstrate that the proposed hybrid optimization based on genetic algorithms can quickly obtain the optimal feature subset and SVM parameters. Therefore the proposed scheme reduces time and improves the accuracy of surveillance video anomaly detection.