农用车辆作业环境障碍物检测方法

Obstacle Detection in the Working Area of Agricultural Vehicle Based on Machine Vision

针对联合收获机视觉导航系统中的视觉测障,提出了一种基于单目彩色图像分割测障与立体视觉特征匹配测障相结合的测障方法:利用H、S颜色分量对单目图像实施固定阈值分割并二值化,获得潜在障碍物的位置及区域;采用尺度空间不变(SIFT)算法获取潜在障碍物区域特征;采用近似最近邻分类算法(ANN)进行快速特征匹配,获得潜在障碍物的世界坐标,由此进一步确认障碍物以及障碍物与车辆之间的距离。提出了提高算法效率的措施,分析了图像压缩比与运行时间、SIFT特征数以及匹配数之间的关系。试验表明,在有障碍物的情况下,检测时间不超过200 ms。

Obstacle detection is a key component of autonomous systems.A obstacle detection method combined monocular vision and stereo vision is studied for the vision navigated combine harvester.For monocular vision detection,H and S components are used to segment the image acquired by the left camera mounted on the combine harvester,and then through the fixed threshold value and binary processing the potential obstacle area is located.For stereo vision,the SIFT features are extracted from the potential obstacle area,and the ANN algorithm is utilized to get matching points.According to the obtained world coordinates the obstacle and the distance from the vehicle are calculated.In order to reduce the processing time the coefficient of image size linear transform is analyzed and it shows that the matching points are enough to satisfy the system need and the processing time is less than 200 ms when the coefficient is 4.0.The experiment using various mature wheat videos indicates that the method is valid to detect obstacles in front of the vehicle.