Control a Drone Using Hand Movement in ROS Based on Single Shot Detector Approach

These days, drones become one of the most popular robots and have lots of application in delivering, transportation, construction, agriculture, data recording. Nevertheless, the increment in using drones, the rate of world aviation accidents increased too, and most of them were because of hardness to keep the drone in wanted location, and also superficial acknowledgement about multi rotors structures. In this paper, a new method is introduced for controlling a drone rather than controlling the drone by classical approach via radio control. In this new method, a deep learning algorithm and machine vision applied to detect an object and use its position to order the drone. In the performed experimental study, which was done in Human and Robot Interaction Laboratory, a human hand was used as an object (to simplify the problem) which its movement is made equivalent to the command of a radio controller. In order to accurate the result of hand detection the so-called Single Shot Detector is used which has been trained by using the Ego Hand dataset. The latter leads to a precise result under different conditions like viewpoint variation, background clutter, illumination and so on. Furthermore, a stereo camera vision set up used instead of using Kinect or leap motion devices to evaluate the depth of hand which enable to control the drone in a three-dimensional coordinate system. Also, the proposed algorithm are implemented in the Robot Operation System(ROS) and Gazebo were applied to get these positions and simulate the drone. The drone, which is based on px4 as a flight controller-in the Gazebo environment- receives its position from Mavros node which connected the drone to the ROS and could move continuously along x, y, and z axes. According to the results, it can be inferred that the proposed method was much easier and had more functionality instead of previous methods reported in the literature.