Research On Multi-channel Video Face Detection And Recognition Technology For Internet Of Vehicles

In recent years,with the rapid development of deep learning technology,face detection and recognition technology has been widely used in many important scenarios.This paper is mainly for the needs of the Internet of Vehicles application scenario.It fully considers the practical challenges of edge node computing resource limitation and real-time processing.This article is mainly for autonomous driving scenarios and future Internet of Vehicles design.It fully considers the real-world challenges such as edge node computing resource limitation and real-time processing.It focuses on lightweight real-time face detection and recognition related algorithms.It also completes embedded system design and implementation.This research will help improve the intelligent monitoring and target discovery capabilities of the Internet of Vehicles.It has important theoretical significance and practical application value.The main research contents of this paper are as follows.This paper fully considers the problem of miss detection/error detection caused by non-ideal image acquisition angle and complex acquisition environment in the Internet of Vehicles scenario.This paper deeply studies the face detection algorithm of MTCNN(Multi-task Cascaded Convolutional Networks).It proposes an improved nonmaximum suppression algorithm.The algorithm combines a small-scale face detection network based on context information.It constitutes a multi-branch cascade neural network.It improves the detection accuracy of tiny faces.After experiment and analysis,the tiny face detection algorithm designed in this paper effectively improves the face detection accuracy in complex Internet of Vehicles networking environment.This paper designs a face recognition algorithm based on face depth features.In order to cope with the problem of severe computing resources limitation on the edge of the Internet of Vehicles,this paper optimized the lightweight network model Lightened CNN for face depth feature extraction.By the batch normalization mechanism,the vector distribution of facial features of different identities is dispersed.The relative feature center distance is increased.It gives the extracted features a better classification aggregation feature.According to the design flow of face recognition algorithm,the method of face alignment and similarity measurement is studied.It is designed to complete the face recognition scheme adapted to the Internet of Vehicles scene.After experiment and analysis,the face depth feature extraction network designed in this paper occupies less embedded system resources.At the same time,the face recognition algorithm designed by combining lightweight face feature extraction network.Face alignment and similarity measurement has higher accuracy.This paper implements and verifies the multi-channel video face detection and recognition system for the Internet of Vehicles on the embedded development platform.The system adopts a modular design concept involving five key modules.Finally,the system will be installed on the embedded development platform.It makes full use of embedded resources through methods such as data sharing between modules,video scheduling,and task management.And it simulates the Internet of Vehicles scenario for system verification.The experimental results show that the face detection and recognition system designed in this paper has good performance.It can meet the application requirements of the Internet of Vehicles scenario in terms of accuracy and real-time detection and recognition.