基于Android平台的障碍物检测系统的设计与实现  

Design and Implementation of Obstacle Detection System Based on Android Platform

作　　者：龚少庆 魏彤[1] 赵剡[1] GONG Shao-qing;WEI Tong;ZHAO Yan(School of Instrumentation Science and Opto-Electronics Engineering,Beihang University,Beijing 100191,China)

机构地区：[1]北京航空航天大学仪器科学与光电工程学院

出　　处：《科学技术与工程》2019年第16期262-266,共5页Science Technology and Engineering

基　　金：北京市科技计划(Z151100002115022)资助

摘　　要：为了提高视觉导盲仪障碍物检测系统的便携性,建立了基于Android平台的障碍物检测系统。对该系统所采用的Android平台采集双目图像、路面平面提取和障碍物检测等算法进行研究。首先,根据Android平台的USB HOST API介绍了Android平台以非ROOT的方式采集双目图像数据;然后,提取并匹配双目图像的特征点得到稀疏的三维点云,在三维点云中用随机抽样一致性算法(random sample consensus,RANSAC)来提取路面平面;最后,在利用双目图像路面平面单应性来区分路面与障碍物的基础上,说明了采用半全局块匹配(semi-global block matching,SGBM)进行障碍物检测的算法。实验结果表明:系统5 m以内的障碍物检测准确率达到90%;检测时间达到1s/帧。完全满足视觉导盲仪障碍物检测系统的检测准确率高、实时性高、易便携等要求。In order to improve the portability of the obstacle detection system of blind guide,an obstacle detection system based on the Android platform was established.The algorithm of external binocular camera,road surface extraction and obstacle detection for the Android platform used in the system was studied.Firstly,according to the USB HOST API of the Android platform,the method of externally connecting the binocular camera to the Android platform in a non-ROOT manner was introduced.Then,a sparse three-dimensional point cloud was obtained by extracting and matching the feature points of the left and right images,and random sample consensus (RANSAC) was used to extract the road surface plane in the three-dimensional point cloud.Then,based on the homography of the road surface of stereo image to distinguish the road surface from the obstacles,the algorithm for detecting obstacles using semi-global block matching (SGBM) was described.The experimental results show that the detection accuracy of obstacles within 5 m of the system reaches 90%,the detection time reaches 1 s per frame.It basically meets the requirements of high detection accuracy,high real-time performance and easy portability of the blind obstruction obstacle detection system.

关 键 词：ANDROID平台 随机抽样一致性算法 单应性 半全局块匹配 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]