基于AI的图像识别乘驾安全监测系统研制  

Research and fabrication of image recognition and driving safety monitoring system based on AI

作　　者：苏明[1] 艾海明[1] 马琳 颜世峰 王东明 SU Ming;AI Haiming;MA Lin;YAN Shifeng;WANG Dongming(Faculty of Science and Technology,Beijing Open University,Beijing 100081,China;Beijing Zhongguancun Science&Technology Service Co Ltd,Beijing 100081,China;Shenzhen Neo Science&Technology Co Ltd,Shenzhen 518000,China)

机构地区：[1]北京开放大学科学技术学院,北京100081 [2]北京中关村科技服务有限公司,北京100081 [3]深圳市尼欧科技有限公司,广东深圳518000

出　　处：《传感器与微系统》2023年第8期85-87,91,共4页Transducer and Microsystem Technologies

基　　金：北京自然科学基金—北京教委联合重点资助项目(KZ201951160050);北京市教委科技项目(KM202251160001)。

摘　　要：利用车载摄像头实时采集车内人员图像信息,自动捕捉车内移动目标,基于人工智能(AI)的图像智能行为识别算法抓取异常动作,如驾驶员异常离开驾驶位、乘客位(副驾、后座)纠纷拉扯、重大交通事故等。系统借助图像智能识别技术,将异常状况信息发送给系统后台,安全监管平台和公安报警系统可通过系统后台与系统主机进行通信,并与事故车辆进行联系。模拟实验采用3大类、15小类,共200项行为动作和器具图像作为乘驾安全监测用例。结果表明:智能图像识别监测准确率不小于96.9%,系统监测实验结果与实验对象真实报警结果经卡方检验无显著差异(P>0.05)。Vehicle-mounted camera is used to collect real-time image information of personnel in vehicle,automatically capture moving targets in vehicle,and capture abnormal actions based on artifical intelligence(AI)image intelligence behavior recognition algorithm,such as the driver’s abnormal departure from the driver’s seat,disputes&pulling in the passenger seat(co-driver seat and rear seat),major traffic accidents,and so on.With the help of image intelligent recognition technology,the system sends abnormal condition information to system backend platform.Security supervision platform and public security alarm system can communicate with the system host PC through system backend platform and contact with the aecident vehicle.Simulation experiment uses a total of 200 behavior actions and appliance images in three categories and 15 sub categories as driving safety monitoring cases.The reault shows that accuracy of intelligent image recognition and monitoring of the system is not less than 96.9%.Chi square test show that there is no significant difference between the system and the real alarm monitoring results of the experimental object(P>0.05).

关 键 词：人工智能 网约车 智能图像识别 卷积神经网络 安全监测系统 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] R318[自动化与计算机技术—控制科学与工程]