Research On Key Technologies Of Ground-Based Robot Collaboration Based On Vision

This thesis studied the vision-based collaborative technology of UAV and unmanned vehicle robots.The topic is divided into three parts,panoramic image mosaic based on SIFT and RANSAC,aerial robotic collaborative three-dimensional scene reconstruction with UAV and unmanned vehicle robots,unmanned aerial vehicle landing detection technology.First of all,image processing techniques are studied,including image preprocessing,corner detection algorithm and camera imaging technology.A variety of image grayscale and image denoising techniques are listed and compared Harris corner detection algorithm and SIFT feature.The algorithms are compared and the experimental results of two algorithms are compared.The relationship between the coordinate system and each coordinate system in the camera imaging model is studied again.Secondly,the panoramic image stitching technology based on SIFT and RANSAC is studied.The basic process is to obtain image sequence,corner detection,feature point matching and image fusion.The SIFT algorithm is used to detect the feature points.Then the SIFT matching algorithm based on the nearest neighbor algorithm is used to initial match the feature points.The method is simple and fast in operation,but it will result in mismatch.Then,the RANSAC algorithm is used to accurately match and improve The accuracy and speed of matching are matched.Finally,two images are mapped to an image by using the image fusion technology to complete the seamless splicing.The fusion algorithm broadens the field of view and improves the resolution of the image.Then,the UAV and ground UAV collaborated to complete the 3D scene reconstruction task.Since UAVs can only get near-surface image sequences,UAVs can be used to assist them in acquiring image sequences at higher and broader locations.After the reconstructed scene was captured,the sparse point cloud was obtained using the Structure from Motion(SFM)algorithm and the dense point cloud was obtained using the Cluster Multi-View Stereo(CMVS)algorithm.Finally,Poisson Surface reconstruction algorithm makes the model surface smoother and more realistic.After the reconstruction work is completed,the characteristics of the scenery can be observed from multiple perspectives,and the two-dimensional plane image can be restored to a three-dimensional structure.Thirdly,the testing of UAV landing mark is completed.In unmanned vehicles to set "T" type of cooperation objectives for UAV identification,and then through the image preprocessing,image segmentation,edge detection,corner extraction and other image processing,to be aircraft relative to the "T" type of marker Key points pose information,through the aircraft landing control system,guide it to unmanned aerial vehicle charging,in order to perform tasks for a long time.Finally,the thesis summarizes the completed work and gives the prospect of the follow-up work.