Design Of Elevator Vision System Based On Face Recognition

Face recognition is an important part of computer vision.In recent years,with the rise of artificial intelligence and the rapid development of face recognition based on deep learning,various algorithms based on deep neural network have been proposed one after another.Compared with traditional face recognition algorithms,they have many advantages,and in some cases,the recognition accuracy has reached or even exceeded human recognition ability.Compared with other biometric technologies(such as fingerprint recognition,iris recognition,etc.),face recognition has many advantages such as naturalness,friendliness,high cost performance,etc.,so it is applied to more and more fields and scenes.This paper focuses on the design and implementation of an intelligent elevator vision control system based on face recognition.The application scenario of the system is set for the buildings with relatively fixed user groups,such as residential buildings,apartments,office buildings,etc.the intelligent elevator system can get the basic information of registered users through face recognition.When the user comes home from work or goes to work,approaching the elevator,the elevator door will open automatically and send it to the corresponding floor according to the identified user information.Its vision control system is to integrate the information from the video screen and the state of the elevator to send out the control command to the elevator,the specific research is as follows:The application scenarios,advantages,social acceptability and feasibility of intelligent elevator system are analyzed.The requirements of normal operation of the vision system are analyzed,and the system is divided into five modules according to the demand information.The docking of vision system and traditional elevator control system is analyzed.The zed binocular camera is selected as the video monitoring input,and its imaging principle,geometric model,camera distortion and correction method are introduced.After comparing three camera calibration algorithms,the traditional calibration method is used to calibrate and correct the camera.This paper analyzes the imaging principle and ranging method of binocular vision,which is used for ranging pedestrians in the operation of the system.Experiments are designed to verify the accuracy of the ranging method.The structure and training method of convolutional neural network are studied to build the model of face detection and face feature extraction.Compared with the current typical target detection algorithms,YOLO V3 is selected for face detection,and its original network model is improved.Then,it is trained on the wide face data set to test the performance of the model.Several classical face recognition algorithm models are analyzed and compared,and facenet is selected for face recognition.Considering that the effect of its original model on Asian face recognition is not ideal,this paper does transfer learning on CASIA-FaceV5 data set,adopts triple loss training model,and tests the model.Design the overall operation process of the system,complete the overall construction of the visual system.The user database is designed and built,and the user occlusion and misidentification problems during the system running are analyzed and optimized.The field video screen is collected to simulate the operation effect of the system.