Research And Application Of Face Recognition Technology In Vehicle Scene

With the continuous expansion of the market scale and the growing number of users,the safety of online car travel has gradually become a social focus of attention.Nowadays,face recognition technology has played an indispensable role in the field of public security.However,most of the current face recognition technologies only have high recognition accuracy in the controlled environment.In the application of real scenes,the accuracy of face recognition will be greatly reduced due to the restrictions of attitude changes,illumination conditions,facial expressions or occlusion.In this paper,we focus on face recognition in vehicle scene to analyze and study the changes of face pose and illumination in vehicle scene.By investigating the common multi-pose variable illumination facial data sets,2000 Asian face image data sets suitable for the application scenarios in this paper are established after screening and integration.In this paper,a face recognition model is constructed based on the combination of directional gradient histogram and gradient lifting decision tree and FaceNet network.Firstly,the direction gradient histogram is used to normalize the input image,and the gradient direction of each cell pixel is calculated and connected into block feature vectors,which are used as feature descriptors to obtain HOG feature vectors to retrieve the face position in the image.Then,68 feature points are obtained from the retrieved face input gradient regression decision model,and gradient regression and affine transformation are performed to verify the results.Face feature points are aligned in different postures.Then 128 feature dimensions are obtained by using the convolution neural network in FaceNet and trained by using the triple loss function to make the features approach the target face gradually.Experiments show that this method can achieve better recognition effect in the face data set set of scene setting.At the same time,the construction of data set and the selection of methods make the model more positive in the actual scene application of face recognition technology.