Research On Binocular Vision Object Detection And Ranging For UAV

In recent years,with the development and maturity of UAV technology,UAV has been widely used in military and civil fields.In addition to carrying out traditional aerial photography,mapping,battlefield reconnaissance and other tasks,UAVs are also given autonomous flight,object detection and recognition functions.Intelligent autonomous UAVs are increasingly welcomed by all walks of life.Binocular stereovision technology and object detection technology are the key technologies of intelligent UAV.They can give the UAV intelligent perception of the environment and improve the autonomy of the UAV.In this thesis,UAV is used as an experimental platform to analyze and design the binocular vision object ranging and object detection of UAV.The distance measurement based on binocular vision is designed,and the disparity and distance information of the object in scene images are obtained by stereo matching algorithm.Firstly,stereo calibration and stereo correction are introduced,and the binocular camera used in this thesis is calibrated.Secondly,several different stereo matching algorithms are introduced,and the advantages and disadvantages of different algorithms are analyzed through experiments.Finally,the stereo matching algorithm is used to obtain the distance information of objects in real scenes.An object detection method based on the combination of MobileNet and YOLOV3 is designed to detect and recognize objects in scene images by deep learning method.Aiming at the problem of low computing power of UAV platform,the YOLOV3 object detection algorithm and lightweight convolution neural network MobileNet are studied in this thesis.The YOLOV3 object detection model based on MobileNet network is designed.The model is trained and tested by GPU.Compared with the original model,the FPS of the model is increased by 3.6 times,and the parameters are reduced by 91%.It is beneficial to deploy the model on embedded devices with less computation.The overall framework of object detection and ranging system based on UAV binocular vision is designed and deployed on the embedded computing platform of UAV.Firstly,the disparity map of scene image is obtained by binocular stereo vision technology;then,the disparity information of the object can be obtained by combining the position information and category information of the object obtained by the object detection algorithm;finally,the real distance of the object can be obtained according to the principle of binocular ranging,thus realizing the detection and ranging of the objects by the UAV.