Research Of Face Occlusion Detection Based On Video Surveillance

With the rapid development of artificial intelligence,intelligent product is becoming more and more universal.Intelligent video surveillance system(IVS)with recognition and processing of visual information,has been received more and more attention.It uses computer vision,pattern recognition and image processing technology to analyze and process video data,and greatly enhance the efficiency of video surveillance system.Recently,face occlusion detection has become a research hotspot in intelligent video surveillance system,which has very important application value in security field.The thesis is to study face occlusion in video surveillance applications,focusing on the issue of the occlusion judgement method.Firstly,moving object detection is studied,real-time foreground-background segmentation using codebook model is analyzed,and head detection based on HOG feature is adopted.Because of the high computational complexity of the algorithm,the real-time performance of video surveillance system can not be met.The head detection based on CUDA parallel computing is used to solve the above-mentioned problem and the effectiveness of the method is verified by experiments.Secondly,a method based on skin color determination and CNN-RBM model is proposed which is based on the head detection algorithm.Considering the poor quality of the head images obtained in video surveillance system,the thesis adopted a method of head images quality assessment(HQA).The method can evaluate the quality of the obtained head image and select some of the higher quality head images.In the skin-based face occlusion detection methods,it is necessary to compute skin color ratio threshold and then determine the face occlusion based on the threshold.Face occlusion detection based on CNN-RBM model is to adopt transfer learning to solve the problem of the few samples on self-built dataset,then use CNN-RBM model which is trained successfully.Experiments show that the two methods have good performance.Finally,a video-based face occlusion detection prototype system is developed in this thesis.At the same time,from the test of the system,it can be seen that the real-time detection of face occlusion in video surveillance system is realized.