一种安保机器人车底高危目标检测方法  

High-Risk Target Detection Method for Security Robot Vehicle Bottom

作　　者：高春艳[1] 吕盛璠 吕晓玲[1] 孙凌宇[1] GAO Chun-yan;LV Sheng-fan;LV Xiao-ling;SUN Ling-yu(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300130,China)

机构地区：[1]河北工业大学机械工程学院,天津300130

出　　处：《机械设计与制造》2022年第8期282-286,共5页Machinery Design & Manufacture

基　　金：国家重点研发计划资助(2018YFB1305301);河北省应用基础研究计划重点基础研究项目资助(17961820D)。

摘　　要：安保机器人作为人工智能、自动控制等技术的综合体,对保障民生安全具有重大意义。受限于车底结构非完整平面,光照不充分,车速非匀速等因素,目前安保机器人对车底高危目标识别准确率较低,检测效率亟待提高。为此,本文提出了一种安保机器人车底高危目标检测方法。首先采用双边滤波器处理输入图像并进行一次小波变换,再将小波域中的低频图像作为输入并利用SURF(Speeded up Robust Features)提取特征点,然后使用基于方向和尺度约束的单向匹配策略寻找匹配点对。在此基础上,采用改进的减小迭代次数的RANSAC(Random Sample Consensus)算法剔除错误匹配点对,并校正目标图像,最后采用基于卷积的NCC算法分块匹配策略寻找危险品位置。实验结果表明,在特征匹配部分本方法相对于SURF+RANSAC算法的匹配正确率提高1.3%,耗时缩短24.2%,在后续目标检测阶段能够标定出危险品位置,满足车底危险品检测要求。As a complex of artificial intelligence,automatic control and other technologies,security robot is of great significance to the security of people’s livelihood.Limited by the incomplete plane of the vehicle bottom structure,inadequate light,uneven speed and other factors,the current security robot recognition accuracy of high-risk targets on the vehicle bottom is low,and the detection efficiency needs to be improved.For this reason,this paper proposes a detection method of high-risk targets on the vehicle bottom of security robot.First,the input image is processed by two-sided filter and wavelet transform is carried out.Then the low-frequency image in wavelet domain is taken as input and the feature points are extracted by SURF(Speed up Robot Features),and then the matching point pairs are found by one-way matching strategy based on direction and scale constraints.On this basis,the improved RANSAC(Random Sample Consensus)algorithm is used to reduce the number of iterations to eliminate the wrong matching point pairs and correct the target image.Finally,the NCC algorithm based on convolution is used to find the location of dangerous goods.The experimental results show that in the feature matching part,the matching accuracy of this method is 1.3% higher than that of SURF+RANSAC algorithm,and the time-consuming is 24.2% shorter.In the subsequent target detection stage,the location of dangerous goods can be calibrated to meet the requirements of the vehicle bottom dangerous goods detection.

关 键 词：安保机器人 目标检测 图像处理 图像匹配 

分 类 号：TH16[机械工程—机械制造及自动化] TP249[自动化与计算机技术—检测技术与自动化装置]