Abstract: The combine of machine vision and robotics is a major trend in the future development of the robotics industry. In the schemes of mobile robot obstacle avoidance, there are many problems with using traditional sensors, and the acquired information is limited. So a method of monocular vision-based mobile robot obstacle avoidance and navigation is proposed. And if the camera lens focal length is known, there is no need for camera calibration in the application. To reduce the impact of light on the edge detection of the obstacle, the color image is converted to the HSI space. Detecting the edge of converted components by using canny algorithm, and the test results are synthesized. Filtering synthetic results by thresholding, to remove weak edge information and improve the detection accuracy. Connecting the spurious edges by morphological processing, and non-obstacle area is obtained through regional growth. The mapping between the image coordinate system and the robot coordinate system is established according the geometric relationship. Finally, using fuzzy logic combined with membership functions to obtain robot control parameters. Experimental results show that the image color space conversion reduces the impact of the shadow and reflective of ground surface, the algorithm can effectively eliminate the interference of ground stripes and accurately detect the edge of the obstacle, and fuzzy logic decision method improves the robustness of the algorithm and the reliability of results.

Key words: monocular vision, mobile robot, navigation, obstacle avoidance, regional growth, fuzzy logic

摘要： 机器视觉与机器人的结合是未来机器人行业发展的一大趋势。在移动机器人的避障导航方案中，使用传统的传感器存在诸多问题，且获取的信息有限。提出一种基于单目视觉的移动机器人导航算法，在算法应用中，如果使用镜头焦距已知的相机，则无需对相机标定。为降低光照对障碍物边缘检测的影响，将机器人拍摄的彩色图像转换到HSI空间。采用canny算法对转换后的分量分别进行边缘检测，并合成检测结果。通过阈值处理过滤合成边缘，去除弱边缘信息，提高检测准确度。采用形态学处理连接杂散边缘，通过区域生长得到非障碍区域，并由几何关系建立图像坐标系与机器人坐标系之间的映射关系。利用结合隶属度函数的模糊逻辑得出机器人控制参数。实验结果表明，对图像颜色空间的转换降低了地面反光、阴影的影响，算法能有效排除地面条纹等的干扰并准确检测出障碍物边缘，而模糊逻辑决策方法提高了算法的鲁棒性和结果的可靠性。

关键词: 单目视觉, 移动机器人, 导航, 避障, 区域生长, 模糊逻辑