Design Of Unmanned Aerial Vehicle (UAV) Automatic Tracking System Based On Qualcomm Platform

Unmanned aerial vehicle(UAV)has the advantages of excellent maneuverability and flexibility which is capable to replace humans do some special task in the complex environment.UAV equipped with camera,using computer vision algorithms to achieve specific tasks has become the research hotspot.However,there are still many problems to be solved,such as limited resources of UAV’s on-board processor,commonly used algorithms require high processor resources,it is difficult to achieve real-time on the UAV.In this thesis,due to limited processor resources of drone,and the intensive computation of commonly used algorithms,we proposed a drone platform based on the Qualcomm platform,using Snapdragon 801 DSP core to achieve precise timing flight control,using ARM core to run Linux system,on which the deployment of OpenCV,and ROS system.A lightweight moving object detection algorithm and a motion tracking algorithm are proposed.All the algorithms are run on the Qualcomm platform in real time.The main work and achievements of this thesis can be summarized as follows:(1)Drone platform construct: This thesis puts forward a small-sized quadrotor hardware structure based on Qualcomm platform and uses DSP core of Snapdragon 801 processor to realize PX4 open source flight control,uses ARM core to achieve the Linux,ROS,OpenCV software environment and related algorithms software architecture,greatly improving the processing ability of UAV platform,reducing the development difficulty,while enhancing the UAV system scalability.(2)Algorithm research of moving object detection: This thesis presents a method of fast extracting the feature points in the video frame using the ORB algorithm.The local difference binary algorithm LDB is used to construct the image descriptors.The nearest neighbor algorithm KNN is used to match the input image descriptor with the target template descriptor,thus a moving object is found in the image.This method has a low computation and requires low ability of processor which is fit for the limited performance of the processor on the mobile platform.(3)Algorithm research of moving object tracking: In this thesis,a method is proposed to fuse the IMU data,GPS information and moving object detection results,and combined characteristics with monocular vision and the attitude of UAV.The relative position of UAV and moving object’s "center of mass" is calculated by using the relationship of coordinate systems between them,and the direction and velocity of the object motion are estimated to realize the algorithm of moving object tracking.The proposed method solved the problem of lacking of depth of view on monocular vision,thus,the relative distance between drone and target object can be calculated.