Research On The Distributed Massive Video Face Recognition Technologies

With the large-scale deployment of monitoring equipment,the massive video data generated by these devices is of vital importance to social and public safety.In big data age,it has always been difficult to find a target from the video through human to see the videos.Based on the current big data processing platform,we design and implement an efficient framework for mass video face recognition.Combined with the application of face recognition,a fast reading strategy of massive video data is designed,and the feature data are optimized in order to enhance the speed of retrieval of facial feature.The main work of this paper has the following four aspects:1.In this paper,aiming at the application features of massive video face recognition,we design a face extraction-retrieval framework based on distributed platform,which includes the storage of massive video,the processing of massive video,the storage and management of massive face images and features.2.In order to sovle the problem that the video can not be sliced at random,the VTI decoupling model of video data is proposed,and the VTI decoupled video data reading in the distributed environment is designed in this paper.At the same time,the problem of data expansion caused by decoupled video is optimized by selecting the key frame strategy,which reduces the information redundancy of the decoupled data.3.Two optimized facial feature extraction methods are proposed: one is to compress the LBP feature from the LBP feature,and the other is to extract the hash code feature from the depth convolution neural network.Through the above two methods to fully reduce the storage space occupied by features,which greatly enhance the speed of retrieval of facial feature space.4.Finally,we implement the massive video face recognition framework on Spark platform,and test the usability and expansibility of the framework.