Research And Application Of Personal Identification Based On Convolutional Neural Network

Identification technology has played an important role in safeguard system,anti-terrorism and other public security fields.However,some traditional identification methods have obvious limitations in practical application.Fingerprint recognition requires that the identified person must cooperate.Face recognition requires a sufficiently clear face image of the identified person.This paper presents a multi-network personal identification method based on multi view human features.Different from the method based on the local features of human body,this method is based on the global feature of human body.The image of different views is trained by using multiple convolution neural networks.Each view gets a CNN model,and then the multiple models are combined by different weights.In the case that a clear facial image could not be obtained,such as the face is blocked or the image resolution is low,especially in the video remote monitoring and other non-contact identification scene,this method shows higher accuracy and stronger robustness.In video surveillance system,it is often desirable to identify people in a more friendly way,such as a remote non-contact.This paper describes how to identify the human target in the surveillance using the multi-network personal identification method based on multi view human features.And the method has achieved the best results in many comparative experiments.This paper describes how to detect and identify human targets in video sequence by using the multi-network personal identification model based on multi-view global features.The detection process is carried out according to the seven steps of obtaining the image sequence,background modeling,background difference operation,binaryzation,open operation,connected domain analysis and extraction of human target.The recognition process is carried out in three steps of constructing data set,training model and identification.Four experiments were conducted to compare the recognition accuracy of different models in a multi-view test set.The results show that the multi-network personal identification model based on multi-view human features has achieved the best experimental results.