Pix2Pix-Based Depth Estimation from Monocular Images for Dynamic Path Planning of Multirotor on AirSim
Pix2Pix-Based Depth Estimation from Monocular Images for Dynamic Path Planning of Multirotor on AirSim
复制标题
DOI：
--
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Tomoyasu Shimada;Hiroki Nishikawa;Xiangbo Kong;Hiroyuki Tomiyama
中科院分区：
文献类型：
--
作者：
Tomoyasu Shimada;Hiroki Nishikawa;Xiangbo Kong;Hiroyuki Tomiyama
最近，对多旋翼的自动武器进行了积极研究。通过深度摄像机测量的距离取决于相机的性能，高性能相机太重且昂贵一种通过轻度和可观的单眼摄像机生成深度映射的方法解决这个问题，本文提出了一种动态路径计划方法，仅使用单眼摄像机避免碰撞，并使用AIRSIM进行模拟提出的方法完全避免了碰撞。
Recently, autonomous flight for multirotor is actively researched. It is essential to get information about the real world for an autonomous flight from sensors. The sensors used by autonomous flight include a camera with a depth sensor or a stereo camera to generate a depth map. A distance measured by a depth camera depends on the performance of the camera. However, the high-performance camera is too heavy and expensive to load on a multirotor. Therefore, this paper proposes a method to generate depth maps by a monocular camera that is light and affordable. Besides, there are legally many limitations on an actual flight of a multirotor, and thus, autonomous flight in virtual environments is usually conducted in an early phase of development. To address this issue, this paper proposes a dynamic path planning method with collision avoidance using a monocular camera only and conducts simulation using AirSim. Experimental results show that the proposed method perfectly avoids collision.