Intuitive 3D Maps for MAV Terrain Exploration and Obstacle Avoidance
Intuitive 3D Maps for MAV Terrain Exploration and Obstacle Avoidance
复制标题
DOI：
10.1007/s10846-010-9491-y
复制
发表时间：
2011-03-01
影响因子：
3.3
通讯作者：
Siegwart, Roland
中科院分区：
文献类型：
--
作者：
Weiss, Stephan;Achtelik, Markus;Siegwart, Roland
最近的发展表明，微型飞行器 (MAV) 如今能够仅使用一个摄像头作为外感传感器，在一个地点自主起飞并在另一个地点着陆。然而，在飞行和着陆阶段，MAV 和用户对整个地形和潜在障碍物知之甚少。在本文中，我们展示了一种实时密集 3D 地形重建的新解决方案。这可用于高效的无人 MAV 地形探索，并为标准自主避障算法和路径规划器提供坚实的基础。我们的方法基于场景的稀疏 3D 点特征上的纹理 3D 网格。我们使用与构建 3D 地形重建网格相同的特征点在 3D 空间中定位和控制车辆。这使我们能够实时重建地形，而无需大量额外成本。实验表明，MAV 可以轻松引导穿过未知的、无法使用 GPS 的环境。在迭代构建的 3D 地形重建中可以识别障碍物，从而很好地避免障碍物。
Recent development showed that Micro Aerial Vehicles (MAVs) are nowadays capable of autonomously take off at one point and land at another using only one single camera as exteroceptive sensor. During the flight and landing phase the MAV and user have, however, little knowledge about the whole terrain and potential obstacles. In this paper we show a new solution for a real-time dense 3D terrain reconstruction. This can be used for efficient unmanned MAV terrain exploration and yields a solid base for standard autonomous obstacle avoidance algorithms and path planners. Our approach is based on a textured 3D mesh on sparse 3D point features of the scene. We use the same feature points to localize and control the vehicle in the 3D space as we do for building the 3D terrain reconstruction mesh. This enables us to reconstruct the terrain without significant additional cost and thus in real-time. Experiments show that the MAV is easily guided through an unknown, GPS denied environment. Obstacles are recognized in the iteratively built 3D terrain reconstruction and are thus well avoided.