Study On Abnormal Emergency Detection System Of The Data-driven Video Surveillance

With increasing attention in recent years, Intelligent Video Surveillance is an application field of computer vision science. The video signal is processed and analyzed through computer vision and image processing. Once the image in the vedio surveillance is abnormal, the system sends alerts and provides useful information so that the work of vedio surveillance on abnormal event can be done effectively. So far, further research has been done by scholars home and abroad, to some extent, solving the problem of massive data, sluggish to respond and low-efficiency of detection for the video surveillance. However, there are still problems such as the generality and accuracy of the detection methods going on.Based on the basic data-driven theoretical method, the thesis, making use of vedio frame image in continual time order, does research on the detection methods of the abnormal event. The detection process can be divided into two parts. In the preprocessing period, gray data on the vedio frame image can be collected respectively on each sample and have it divided into N*N, and then withdraw the average gray in each pixel into the feature matrix and reduce dimension of the data. In the abnormal event detection period, the thesis provides two methods:(1)Compare the similarity with normal event to tell whether it is an abnormal event.(2)From the perspective of statistics, according to the dataset of normal event that has been defined previously, set the related threshold and detect whether it is an abnormal event based on Infinite Hidden Markov Model. To ensure the real-time performance of the system, FPGA is adopted as the main controller for its parallel execution and rich logic resources. The image acquisition, the abnormal emergency detection and result display are all accomplished within the same chip of FPGA, which improves the integration of the whole system. All the modules in the system are programmed in Verilog HDL. Thus the processing speed of the system improves significantly. Experiments have shown that this design, namely the abnormal emergency detection system of the data-driven surveillance controlled by FPGA, can be widely used with excellent real-time performance and convenience to update and maintain.