基於物體偵測實現之無人機避障系統

Abstract

無人機相關應用在未來越來越重要。有了無人機的協助，人類的生活將會變得更加便利。目前有許多無人機相關應用，像是監測系統、科學或學術研究、運輸、……等等，這些應用在目前都在人為操控下完成任務。若這些無人機應用能在無人操控之下完成任務，將會省去不少人力。因此，在我們的研究中，我們希望能提出一套讓無人機能自動避障的方法。 在我們的研究中，我們提出一個只需要一台網路攝影機以及1D的LIDAR感測器兩個低成本的配備的避障系統。在這系統中，首先先做物體偵測，利用物體偵測的框框位置資訊，得出每個物體的深度資訊。接著，再用我們提出的避障演算法，使無人機能找到一個可行的避障路徑。整套系統可以運行在嵌入式系統上，因此適合在硬體資源有限的無人機平台上應用。

UAV applications becomes more and more important in the future. Our world might be more convenient with the help of the UAV. So many applications like surveillance, science and academic research, transportation, and so on, can be achieved with the help of UAV. Recently, most of the UAV applications are under the manual control. But if the UAVs can finish the mission without human control, it will save lots of human resource. So in our research, we aim to propose an obstacle avoidance system for UAV. In our work, we propose an obstacle algorithm just using one webcam and one 1D LIDAR which are low cost. We firstly do the object detection and use the bounding box position information to get the depth information of objects. Also, we proposed our obstacle avoidance algorithm to make a feasible route planning decision. The whole system can run on embedded system which is suitable for limited hardware resource UAV application.