Research On Key Technology Of Bionic Hydraulic Quadruped Robot Navigation Based On Stereo Vision

With the improve ment of robot technology and t he expansion of application domain,aut onomous navi gati on technology based on machi ne vision has become a key topic i n the field of robot. To t hose quadruped r obots relying on feet walking mechanism under unstruct ured condition, it is quite difficult t o detect roadblocks i n advance and provide 3D envi ronment i nformation during walking. So that an autonomous naviga tion s ystem with high accuracy, high speed, hi gh reliability and high robustne ss de serves exhaustive studies.On the basis of the bionic hydraulic quadruped robot developed for logistics support of mountain wars, which demands reliable walking and stabl e operation under unstructured envi ronment, model ing, simulation and experiment were combined for the systematic,ordered, deep and deta ile d exploration and r esearch on stereo vision navigation. The prima ry work and achi eveme nts are as follows:⑴ By analyzing requirement s of the st ereo vision naviga tion system, general de sign was fi nished on the basis of r easonabl e funct ions and stable performances. Hardware pla tform of parallel stereo vision sys tem was established ac cording to optical imaging principle and act ual pa rameter requirements.⑵ Image preproc essing algorithm including image acqui sition, grayscale convertion,smoothi ng and s harpeni ng was studied. I ntrinsic and extrinsic para meters of cameras were calibrated, the position and orientation of camera s we r e corrected and epipolar li nes of stereo image s were rectificated.⑶To solve the trouble t hat images collected in fiel d environme nt have large areas with similar grayscale and the bad real-time problem dur ing dense matching proc ess, wat ers hed al gorithm a nd fuzz y c-means clus tering algorithm wer e combi ned for contour extraction,which would lead to good noise immunity, robustness and real-time.⑷ Aiming at the problem of error matching, a double const raint with region matching and pixel matching was propose d for optimization, which guaranteed the speed and accuracy requirements by r educing searching dimensional ity and obtained a smoothing dispari ty agreed with real scene.⑸ On the bas is of binocular stereo vis ion model and triangulat ion principle, the 3D coordinate calculation method of spat ial poi nts could be derived fr om cor responding subpoints in imaging pl anes, which promoted the 3D reconstruction to a high s tandar d.In order to prove the correc tion and validity of the method described in this paper, a binoc ular vision measuring system was built by analyzed parameters, and experiments like cont our extraction, stereo matching and 3D reconstruction were carried out by s imulation software Matlab. The experimental res ults show t hat time for stereo matching have shaved90% since the number of ma tching pixels e xtrac ted by contour feature method we re much le ss than othe rs, meanwhile the accura cy of matching and reconstruction has been advanced by adopt ing double mat ching cons trai nt. Thus algorithm proposed i n this pa per meets the expected des ign requirements and creates necess ary conditi ons for the bionic hydraulic quadruped robot to walk re lia bly and oper ate stably under unstructured environment.