Research On Uav Obstacle Detection Based On Binocular Vision

In recent years,small unmanned aerial vehicles have developed rapidly in the direction of autonomy and intelligence,and the application scope has become increasingly wide,especially in the absence of satellite navigation conditions.Rotor unmanned aircraft development initial period mainly for consumer users to design,not too much to consider its practical application in engineering or may encounter in the scene,the main by manipulation of the hand by the naked eye to discern flight direction,in the case of complex environment that puts forward high demands on operators,even skill is difficult to guarantee the safety of the unmanned aerial vehicle,night at the same time the lack of light,it's difficult to see without external force with the help of the operators of unmanned aerial vehicle itself.The traditional navigation system composed of inertial measurement unit and GPS cannot meet the requirements.This topic is based on binocular stereo vision technology,which USES images obtained from different perspectives to obtain the depth information of images through parallax,thus obtaining the three-dimensional information of the scene.As an important branch of machine vision,binocular stereo vision can obtain the distance information of 3D objects in the scene,it is mainly divided into the camera calibration,image acquisition,image preprocessing,feature detection and extraction,stereo matching and scene distance information acquisition and so on six steps.This paper focuses on the three steps of camera calibration,image acquisition and stereo matching.In terms of camera calibration,adopted ilf two-dimensional plane calibration method,combining with the calibration of the Matlab toolbox,successful calibration respectively about the inside and outside the parameters of two cameras and the rotation of the translation between the two cameras parameters,calibration of send some parameters of high precision,fully meet the needs of binocular stereo vision system of this paper.The acquisition of image is the most basic step in the whole binocular vision system.Improving the performance of image acquisition is helpful to improve the overall performance of the system.On the FPGA design of synchronous signal acquisition module through the camera parallel two cameras for image acquisition,through internal bus,the DAM will be collected images of transmission to the DDR,this kind of method to improve the efficiency of image acquisition and transmission,reduce the burden of the CPU,improve the performance of the whole of the binocular vision system.As binocular stereo vision is one of the most important,one of the key steps,stereo matching,by about two regional gray level characteristics,point in the image or phase characteristics,such as matching,get the match after two projection point,through the principle of triangle calculation to obtain the corresponding scene point distance information.Based on the Harris corner feature detection algorithm and SIFT stereo matching algorithm,this paper realizes the binocular stereo matching by combining these two algorithms.The first to use this method to test the two images around,Harris operator and then to detect the corner generated SIFT feature vector,according to the obtained SIFT feature vector,the Angle points in two images matching,matching point.The method of using this method is less time-consuming,and the matching efficiency and accuracy are higher than that of SIFT alone.