Research On Application Of Facial Recognition Technology In Electronic Locks

With the development of the times,people's trips have become frequent.In many public places,lockers have become an indispensable tool,which solves the demand for people to store items in fragments.At the same time,with the improvement of people's quality of life,the existing lockers can no longer meet people's requirements.Face recognition is gradually used in locker systems due to its uniqueness and non-contact characteristics.Compared with traditional storage methods,facial recognition-based storage methods are safer and more efficient.Therefore,it is of great significance and value to study the locker system based on facial recognition technology.In face detection and recognition tasks,convolutional neural networks show excellent performance.In order to achieve higher accuracy,designing larger and deeper network models has become the main trend,but deeper network models are generally not suitable for embedded platforms with limited computing power and storage space,such as electronic lockers.Therefore,the main purpose of this thesis is to design and implement a lightweight face detection and recognition model suitable for embedded platforms with fast running speed,small model size,high detection and recognition accuracy.This thesis takes the face detection and recognition model based on deep learning as the research object,carefully studies their algorithm ideas and shortcomings,optimizes and improves the model around the purpose of "lightweight",and applies the model to the Raspberry Pi platform to complete The design of the locker system.The main research work of this thesis is as follows:First,in terms of face detection,the SSD?Mobile Net network model is optimized.Firstly,the dilated convolution is used to replace the traditional convolution as the first layer convolution.The dilated convolution can obtain more visual information and detailed features under the same parameters and calculation amount.Secondly,drawing on the lightweight ideas of Mobile Net,optimizing the depth of separable convolution,designing Tiny building blocks and improving the information loss problem of deep separable convolution.Designing a feature fusion network to superimpose the fuse shallow features and deep features.Enriching the feature information to improve the detection accuracy.Finally,the feature detection layer is redesigned to make the face detection model more lightweight.Secondly,in terms of face recognition,the optimized Mobile Net network is used to replace the backbone network Google Net in the Face Net to make the face recognition model more lightweight.Remove the fully connected layer at the end of the network,and adopt a global average pooling layer with fewer parameters and calculations,which greatly reduces the volume of the model and the amount of calculation while ensuring the recognition accuracy.In the model training stage,Center Loss and Softmax Loss are introduced successively to jointly optimize the training network to obtain a high-precision lightweight face recognition model.Thirdly,using the above research results,an electronic locker system based on facial recognition technology was designed and implemented.The human-computer interaction module of the locker system is designed and implemented in detail,and the locker system is applied to the embedded platform for functional testing and performance testing.The experimental results show that the system is stable and reliable,has good real-time performance,and has a good accuracy rate,which meets the basic requirements of actual storage scenarios.