融合视觉图像处理的目标识别技术在移动机器人中的应用  

Application of target recognition technology based on visual image processing in mobile robot

作　　者：潘萍[1] PAN Ping(Changsha Normal University,Changsha 410148,China)

机构地区：[1]长沙师范学院,长沙410148

出　　处：《兵器装备工程学报》2024年第3期314-322,共9页Journal of Ordnance Equipment Engineering

基　　金：湖南省自然科学基金项目(2015JJ6007)。

摘　　要：为了提升移动机器人的运动人体识别效果,研究利用Faster-RCNN-KF对动态人体图像进行实时跟踪,并结合Facenet-MTCNN实现跟踪对象人脸识别。运动人体跟踪检测实验和人脸识别测试结果显示,Faster-RCNN-KF算法跟踪误差仅为0.000 5 m,且跟踪响应速度和误差更正速度较快;Facenet-MTCNN目标识别算法在训练中的分类精度最高能够达到99.15%,分类中的时间延迟为0.01 s,能够有效识别跟踪对象的身份信息。研究结果表明,视觉图像处理技术能够实现人体的有效跟踪检测,并能对不同身份的跟踪对象进行人脸识别,对移动机器人跟踪与识别技术发展具有重要价值。In order to improve the recognition performance of moving human bodies in mobile robots,the study utilizes Faster RCNN KF for real-time tracking of dynamic human body images,and combines Facenet MTCNN to achieve facial recognition of tracked objects.The results of human motion tracking and detection experiments and facial recognition tests show that the tracking error of the Faster RCNN KF algorithm proposed in the study is only 0.0005 m,and the tracking response speed and error correction speed are fast.At the same time,the Facenet MTCNN target recognition algorithm proposed in the study has a classification accuracy of up to 99.15%in training,and a time delay of 0.01 seconds in classification,which can effectively identify the identity information of the tracked object.The research results indicate that visual image processing technology can achieve effective tracking and detection of the human body,and can recognize faces of tracked objects with different identities,which is of great value for the development of mobile robot tracking and recognition technology.

关 键 词：人体跟踪 人脸识别 图像处理 Faster-RCNN 卡尔曼滤波 Facenet网络 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]