CSI-Based Violent Behavior Detection Method

People are deeply saddened by the endless incidents of domestic violence and school violence. However, the existing real-time monitoring methods of violence have some shortcomings, such as high deployment cost, exposure of privacy and so on. In order to solve the above problems, a violence detection method based on channel state information (CSI) is proposed. Firstly, uses wavelet denoising and smoothing filtering to suppress the noise of signal, and then calculates the sliding variance to extract the existence of activities. Secondly, in order to make full use of the effective information in CSI, combine image domain, time domain, and frequency domain features, and finally put the three-domain features into the support vector machine(SVM) for behavior classification. The experimental results show that the average recognition accuracy of classifying 10 kinds of behaviors (daily and violent behavior) in darkroom and laboratory scenes can achieve 97.3% and 92.7% respectively.