An Efficient Real-Time Indoor Autonomous Navigation and Path Planning System for Drones Based on RGB-D Sensor

Abstract

An efficient real-time autonomous drone navigation and path planning system is proposed by using an RGB-D sensor in the GPS-denied indoor environment. Firstly, the RGB images and their corresponding depth images collected by the sensor were used for real-time pose estimation and dense point cloud map reconstruction based on ORB-SLAM (Simultaneous Localization And Mapping system based on the ORiented BRIEF feature). With the attitude angle data from the Inertial Measurement Unit (IMU) used in the pose initialization, the SLAM system can provide a more accurate pose information, and therefore a better 3D map when the drone flies with a high speed. In the second phase, a simple and efficient pathfinding program was developed. In this procedure, the point cloud map data would be translated to a map based on the octree structure, which contains the occupied and free voxels. The size of occupied nodes in the octree map is three times smaller than that of free nodes, which ensures more free area for the drone’s flight. Every free voxel in the octree map was set to have enough space for the drone to cross. An improved $A^*$ algorithm map was used to plan the route based on the octree. The simulation result showed that this efficient pathfinding method can reduce execution time and memory space significantly. In the actual flight, a local path planning system was proposed to avoid obstacles in real time.