Research On Android-based Remote Security Personnel Monitoring System

With the rapid development of artificial intelligence and computer vision technology,autonomous driving has become the mainstream direction for the future development of the automotive industry.However,current trials of autonomous vehicles both domestically and internationally still require onboard safety personnel for vehicle supervision and warning,greatly increasing the cost of commercial operation for autonomous vehicles.Furthermore,while policies have allowed for the transition of onboard safety personnel to remote video monitoring,the effectiveness of remote monitoring for vehicle supervision remains uncertain.Based on this background,this study is based on a "One-Click Unmanned Shared Valet Parking" project by a certain automaker.It analyzes and studies adverse monitoring behaviors and,in conjunction with the practical operational management needs of the OneClick Shared Valet Parking,preliminarily builds a remote autonomous driving monitoring system on the Android platform.This system not only receives vehicle condition-related information but also incorporates relevant judgments of the effectiveness of remote safety personnel monitoring.The goal is to ensure that the safety personnel and vehicle status are continuously monitored in both directions during the low-speed autonomous driving dispatch process,thereby ensuring the safety of the vehicle and pedestrians.The main content and innovations of this research are as follows:1)Proposed an improved Blaze Face facial detection algorithm.Based on a summary and analysis of existing facial detection algorithms,the lightweight convolutional neural network Blaze Face,suitable for deployment on mobile device GPUs,was selected as the backbone network.An attention mechanism called the CBAM module was introduced to improve the accuracy of small target facial detection.The improved algorithm achieved accuracy rates of 90.5%,88.3%,and 77.6% on the Easy,Medium,and Hard categories of the Wider Face dataset,providing a foundation for subsequent adverse behavior feature extraction.2)Through analysis and experimental comparison of existing facial feature extraction methods,it was found that deep learning-based facial feature recognition algorithms can only perform qualitative analysis of facial conditions and have weak generalization capabilities.Pixel feature point distribution-based algorithms relying on the "three court five eyes" principle are susceptible to environmental disturbances such as dim lighting and wearing glasses,leading to significant misjudgments of behavior.On the other hand,the SDM-based facial feature point detection algorithm can accurately and rapidly locate facial feature points and calculate the head’s spatial pose angles using the POSIT algorithm.Experimental results demonstrate that this algorithm can accurately extract safety personnel’s facial and head pose information.3)Considering that adverse monitoring behaviors have multiple features with nonunique feature weights,a method for fusing multiple features,including blink rate,yawning rate,head-down rate,and head anomaly rate,was proposed.Using rough set theory,the judgment results of multiple indicators were fused to obtain the final evaluation result.Experimental results show that the fused adverse monitoring evaluation criteria achieved an accuracy rate of over 93.5% on 1000 randomly selected images from the Yaw DD public video dataset,demonstrating high accuracy.4)Addressing the issue of most monitoring backends being deployed on PCs and the reliance on smartphones for work,this study proposes a solution for deploying a remote monitoring system on the Android platform.This solution can reduce the constraints of fixed workstations for personnel and further reduce the workload of safety personnel.Additionally,a custom data packet structure was designed to address packet concatenation and splitting issues when transmitting data from the onboard Tbox to the server through gateways.Furthermore,to improve real-time interaction between the client and server,Web Socket real-time communication technology was adopted to establish a long connection between the server and client,effectively reducing the latency in client-server interactions.This study researched an Android-based remote autonomous driving monitoring system based on monitoring effectiveness,enabling the judgment of safety personnel monitoring effectiveness and mobile acquisition of vehicle condition information.It provides new ideas and methods for optimizing remote monitoring systems,which are of great significance in promoting the commercialization of unmanned shared vehicles.