Research And Development Of Key Technologies Of Intelligent Stereo Garage System Based On Face Recognition

The intelligent stereo garage is a new engine to solve the parking problems and promote the development of smart parking industry.It is of great theoretical and practical significance to delve the intelligent and informatization control system,and the convenient dentify authentication in the garage scene.With the support of the key R&D project of Jiangsu Province,this thesis researchs and develops the key technologies of intelligent stereo garage system based on face recognition.A dual-mode ARM main control device,which combines intelligent control,data collection and human-computer interaction in one,is researched and developed.Meanwhile,the face detection algorithm for people in cars and the face recognition algorithm for people accessing cars are researched and applied under the environment of stereo garage,to enhance the intelligent and informatization level of the stereo garage,and promote the construction of urban smart parking industry with unattended operation and contactless authentication.Firstly,a face detection algorithm for people in cars based on deep learning is researched and applied in embedded environment.Based on the analysis of MTCNN algorithm,this thesis improves the method of generating image pyramid and proposed a candidate frame prediction network based on multi-scale feature fusion.Meanwhile,the EIo U Loss function,which normalizes multifaceted data,is used to enhance the calibration accuracy and efficiency of the candidate frame.The good performance of the improved algorithm under the environment of stereo garage is verified by comparative experiments.Secondly,a lightweight face recongnition algorithm for people accessing cars based on metric learning is researched.A classical model based on metric learning,called Face Net,is analyzed about the reason for the degradation of the performance under the environment of stereo garage.Aiming at the problem of large differences in facial pose and clarity when accessing cars,an attention module which integrades multi-scale spatial attention and channel attention is designed to strengthen the feature extraction and representation capabilities of the original network.Furthermore,the depthwise separable convolution and the H-Swish activation function are applied to make the model more lightweight and efficient.And a hyper-parameter for reducing of the channels of depth separation convolutions is used to build model with right size for the constraints.The comparative experiments show that the reconstructed network trades off between latency and accuracy efficiently,and meets the application requirements of the garage haredware well.Finally,a main control device based on face recognition for intelligent stereo garage is designed and developed.Accroding to the dual-mode ARM architecture,which includes a core control device based on Cortex-M4 architecture and a self-service terminal device based on Hi3516DV300,the key chips selection and implementation of the system functions are expounded in modules,such as parking scheduling,data communication with NB-Io T,and data collection of the status of garage.Meanwhile,the embedded interactive interface of the terminal device is developed to provide functions such as face authentication and local management.After the joint debugging and testing at the garage scene,the system has been verified with complete functions and the ability to run stably.The core control and face authentication technology of the intelligent stereo garagesystem,researched and developed in this article,sovles the problems of complicated operation and the low level of automation.The relevant performance indicators of the system have reached the requirments of th article.The system has been put into trial operation in January 2022,and improve the level of intelligence,automation and informatization of stereo garage effectively.