基于空地信息互补的无人车路线规划

Route planning of unmanned vehicle based on complementary air-ground information

空地机器人通过相互协作获得的信息互补特性,使其在多智能体协同领域具有显著优势。由于室外场景受地形条件、光照变化等影响,地面机器人凭借自身传感器无法完整获得有效的地面环境特征,因此提出一种室外大范围场景下的地面特征提取方法,可通过获取到的全局环境信息指导地面机器人完成导航任务。利用机载视觉及YOLOv3目标检测算法对室外场景的目标物进行检测与定位;对于全局场景图像特征的提取,建立点对间映射函数模型,剔除待匹配图像的外点,从而配准场景图像并对其进行拼接,使用加权平均融合淡化拼接缝隙;通过k(聚类中心数)均值聚类算法分割全景图像,并采用形态学方法滤除噪声,最终获得全局场景特征;在此基础上改进基于采样的路径规划方法,为地面无人车提供一条可行的全局路线。通过仿真实验及室外场景测试表明:地面场景特征提取算法提升了室外大范围场景下所提取到特征的鲁棒性和准确性,改进的路线规划方法进一步缩短了路径长度,可为无人车自主导航提供有效的全局环境信息。

The complementary characteristics of information obtained by air-ground robots through mutual cooperation have significant advantages in the field of multi-agent collaboration.Due to the limitations of terrain condition and illumination changes in outdoor scenes,the ground robots cannot completely obtain effective ground environmental information with their own sensors.In this regard,a ground scene feature extraction method in large outdoor scenes was proposed to guide the ground robots to complete the navigation task through the obtained global environment information.The airborne vision and YOLO-v3 target detection algorithm were used to identify and locate the target in the outdoor scene.For the feature extraction of the panoramic image,a robust correspondence function model between point correspondences was established to eliminate the outliers during the image registration procedure and thus to stitch the images,thereby achieving a more accurate image registration result.The weighted average fusion method was also applied to dilute the stitching gap.The k(clustering number)-means clustering was employed to segment the panoramic images,along with the morphological method to filter out the background noise and thus obtained the final scene characteristics.On the basis of the global scene features,a feasible global route was provided for the ground unmanned vehicle by improving the sampling-based path planning method.Through simulation experiments and outdoor scene tests,the ground scene feature extraction algorithm improved the robustness and accurateness of the extracted environmental feature under wide range of scenarios and the path length was further reduced by the improved route planning method,thus provided more effective environmental information for autonomous navigation of the unmanned vehicle.