| Live streaming is recently becoming increasingly popular.Applications including online TV programs, online streaming on social media, and real-time surveillance systems boost drastic increase in demand for live streaming. How to efficiently detect abnormal cameras will be crucial to the system availability,especially unmanned surveillance system. In our previous research, a real-time streaming processing platform was developed based on Apache Storm for abnormal event detection of cameras. However, with regard to certain kinds of fault types
such as mosaic frames and blurry frames, it is difficult to tackle by means of traditional image processing methods. In view of the issue, we have proposed an intelligent event detection model based on Convolutional Neural Networks in order to improve the
detection accuracy. To prove the usability of the proposed model,
the performance has been verified by performing experiments on
a live streaming testbed. According to the experiment results, the
prediction accuracy of proposed model could reach
approximately 96.5%. |
