| 引用本文: | 胡钊政,伍锦祥,肖汉彪,周哲.基于逆投影差分的移动机器人障碍物快速检测[J].哈尔滨工业大学学报,2022,54(11):95.DOI:10.11918/202112045 |
| HU Zhaozheng,WU Jinxiang,XIAO Hanbiao,ZHOU Zhe.Fast obstacle detection based on inverse perspective mapping difference for mobile robots[J].Journal of Harbin Institute of Technology,2022,54(11):95.DOI:10.11918/202112045 |
|
| |
|
|
| 本文已被：浏览 869次 下载 912次 |
码上扫一扫！ |
|
|
| 基于逆投影差分的移动机器人障碍物快速检测 |
|
胡钊政1,2,伍锦祥1,2,肖汉彪2,3,周哲2,3
|
|
(1.武汉理工大学 信息工程学院, 武汉 430070;2.武汉理工大学 智能交通系统研究中心,武汉 430063; 3.武汉理工大学重庆研究院,重庆 401120)
|
|
| 摘要: |
| 移动机器人的障碍物快速检测是其导航、避障、轨迹跟踪的关键技术。目前多数传感器存在距离盲区以及异常尺度等问题,且现有算法大多计算复杂,难以满足实时性需求。为了改善这些问题,本文提出了一种基于环视逆投影差分的移动机器人障碍物快速检测与定位新方法。该方法以水平地面为参考平面,对移动机器人上环视系统中相邻视角相机拍摄的图像进行逆投影变换,再经图像差分得到逆投影差分图,图中像素值为零的点位于参考平面,反之则位于参考平面之外,经过阈值化、滤波后即可用来区分参考平面与非参考平面目标,实现其视场公共区域内障碍物的快速检测；同时以逆投影差分图为基础,通过像素坐标系到机器人自身坐标系的转换可获得目标相对于移动机器人精准的位置信息。在实际场景中对5 m范围内不同类型、大小、距离的障碍物进行测试,平均检测精度为97.3%,平均检测耗时为46 ms/帧,平均定位误差为1.1%。实验结果表明,本文方法能快速有效地检测和定位移动机器人附近的动、静态障碍物。 |
| 关键词: 移动机器人 计算机视觉 逆投影 障碍物检测 障碍物定位 |
| DOI：10.11918/202112045 |
| 分类号:TP242.6 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(U1764262)； 重庆市自然科学基金(cstc2020jcyj-msxmX0978) |
|
| Fast obstacle detection based on inverse perspective mapping difference for mobile robots |
|
HU Zhaozheng1,2,WU Jinxiang1,2,XIAO Hanbiao2,3,ZHOU Zhe2,3
|
|
(1.School of Information Engineering, Wuhan University of Technology, Wuhan 430070, China; 2.Intelligent Transportation Systems Research Center, Wuhan University of Technology, Wuhan 430063, China; 3.Chongqing Research Institute of Wuhan University of Technology, Chongqing 401120, China)
|
| Abstract: |
| The fast detection of obstacles is the key technology for the navigation, obstacle avoidance, and trajectory tracking of mobile robots. Most sensors have problems such as distance blind zone and abnormal scale, and the existing algorithms are generally limited by complex calculation, making it difficult to meet real-time requirements. This paper proposes a new method of rapid obstacle detection and localization for mobile robots based on the difference of inverse perspective mapping. The method takes the level ground as the reference plane to perform inverse perspective transformation on the images captured by the adjacent view cameras in the around view monitoring system mounted on the mobile robots, and obtains the inverse perspective mapping difference by image difference. In the images, the point with zero pixel value is located on the reference plane; otherwise it is outside the plane. After thresholding and filtering operation, the method can be used to distinguish targets not on the reference plane, so as to realize the fast detection of obstacles. Besides, obstacles can be located accurately on the basis of the inverse projection mapping difference through the transformation from the pixel coordinate system to the coordinate system of the robot. Different types, sizes, and distances of obstacles within 5 m were tested. The average detection accuracy was 97.3%, the average detection time per frame was 46 ms, and the average localization error was 1.1%. The experimental results show that the proposed method can quickly and effectively detect and locate the dynamic and static obstacles near the mobile robot. |
| Key words: mobile robot computer vision inverse perspective mapping obstacle detection obstacle localization |
|
|
|
|
