Detection of Normal and Abnormal Human Activities for Video Surveillance

As the number of users of surveillance cameras increased in public places, the need was felt that the camera automatically recognize normal and abnormal activities were to quickly achieve higher detection and diagnosis without fatigue and human error can notify the protective forces. For this purpose, we introduce an integrated descriptor where the background and foreground information simultaneously apply Fourier transform to extract the information, we do feature extraction using Gabor filter to reduce the dimensions of a space and feature selection Maximum Relevance Minimum Redundancy (MRMR) use and the training and support vector machine classifier will recognize the activity. The advantages of this approach are the property of not separating foreground from background to meet the needs of Arbitration in Sport games as well and the use of classifiers using feature selection step and the speed and accuracy MRMR to reduce the dimensions noted.