| Now in this hugely populated society,the issue of security is becoming more and more people’s concern.Due to the economic development,people’s living conditions are getting better and better,and the requirements for safety are getting higher and higher.Surveillance cameras are everywhere in the streets and lanes,covering an increasingly large area.Traditional video surveillance features a single,only able to store,capture and playback video,this feature can only be used for recording,it is difficult to play the role of early warning and alarm.If you want to be sure to find the target person in the surveillance video in real time,you need the supervisor to watch the multi-channel surveillance video.Due to the supervisory staff watching the video for a long time,it is extremely easy to be exhausted.Especially when there are many surveillance videos,many important information often missed and it is very difficult to find the target person accurately.For these reasons,an intelligent video surveillance system need to help supervisors to get the job done.In view of these phenomena and problems,this paper proposes the design and implementation of face recognition method in the target-driven surveillance video positioning system.The face recognition method in this paper has the advantages of high accuracy and low computational cost.This dissertation divides the face-recognition method in the target-driven surveillance video positioning system into four modules:video processing module,face detection module,face coding module and face determination module.The face coding module encodes the face image into a 128-dimensional vector in the European space.The coding process is performed by using a convolutional neural network.The face determination module takes the 128-dimensional vector as input,and uses the SVM classifier to determine the 128-dimensional vector to identify whether this 128-dimensional vector belongs to the target face.Then we have a detailed design and implementation of each module.Convolution neural network design of human face coding module draws on the thought of Inception network and Residual network,convolution neural network using TensorFlow framework implementation,and face recognition module support vector machine using Scikit-Learn framework.Finally,based on the convolutional neural network and SVM designed and implemented in this paper,we train it using public datasets and test it for functionality.The experimental results show that the face recognition method designed in this paper can recognize face images with high accuracy and low computational cost,and can realize the function of target-driven video surveillance and positioning system. |
