Research On Environment Perception Method Based On Binocular Stereo Vision

UAVs have the advantages of compact size,high resilience and low cost,and they are used in many areas of society.Routinely,the landing of UAVs are controlled by the flight controller to land in a pre-defined flat area,or controlled by the procedure to independently identify the predetermined ground cooperation target to achieve autonomous landing.As the scope of UAV applications continues to expand,emergency landings are often required when unexpected events happened,which requires the UAV to have the ability to autonomously sense the terrain and identify flat areas based on terrain information,if the flight controller is not familiar with the terrain and there is no ground cooperation target in the landing area.In view of this,in order to enable the UAV to acquire the ability to autonomously perceive flat areas on the ground,this thesis carries out research on the environment perception method based on binocular stereo vision,focusing on the following two problems: first,how to overcome the influence of light and noise in the outdoor environment and acquire dense three-dimensional information on the ground;second,how to seek out the flat areas suitable for UAV landing based on the acquired threedimensional information on the ground.The achieved results include the following aspects.For the problem of high mis-matching rate of census stereo matching algorithm when acquiring dense ground 3D information in outdoor lighting and noise-influenced environment,this thesis proposes a semi-global stereo matching algorithm based on weighted census.In this thesis,we propose a semi-global stereo matching algorithm based on weighted census,which reduces the false matching rate while improving the census transformation that relies too much on the central pixel point.The experimental results show that the average mis-matching rate of the semi-global stereo matching algorithm based on weighted census is lower than that of the census stereo matching algorithm under different noise,different illumination and exposure in the non-occluded area,which improves the matching accuracy and achieves the purpose of obtaining dense parallax images.For the problem of fast and accurate autonomous screening of flat areas suitable for landing by UAVs without the help of ground cooperative targets after the ground 3D information obtained,this thesis proposes a plane fitting algorithm based on parallax image localization.The parallax image is divided into blocks,and a least squares-based plane fitting algorithm is used for each local area to determine whether the local area is flat or not based on the mean square difference and the fitting determination coefficient.The experimental results show that the parallax-based local plane fitting algorithm achieves the ability to perceive the flat areas in the scene more accurately.