Research On Recognition And Docking Positioning Method Of Fuel Tank Interface Based On Binocular Vision

With the development trend of intelligent gas stations in the future,using intelligent refueling robots to replace the traditional manual refueling mode can reduce the number of fuel support personnel and improve the efficiency and safety of refueling.Binocular stereo vision system can help the intelligent refueling robot quickly obtain scene spatial information,which has great advantages in short distance measurement.But due to the complexity of the operation scene and the characteristics of the fuel tank interface,its identification and positioning performance still need to be improved.For the fuel receiving vehicle using the dry disconnect coupling as the fuel tank interface,this thesis studies the identification and docking location method of the fuel tank interface based on binocular vision and completes the identification and docking location of the fuel tank interface in the scene.The main work of this thesis includes:Firstly,this thesis thoroughly studies the imaging principle of the visual system and the conversion relationship of relevant coordinate systems,and establishes a binocular camera model.Then it analyzes the requirements of the intelligent refueling robot visual recognition and positioning system,selects a binocular camera that meets the requirements,and builds a system hardware platform to complete the system software algorithm flow design.Secondly,combined with the characteristics of image acquisition and fuel tank interface,this thesis uses modified alpha filtering and Gamma correction to preprocess the images.Aiming at the problems of false matching and low real-time performance of feature point pairs in the process of fuel tank interface recognition by SURF based target recognition algorithm,the feature descriptor was improved to enhance its rotation invariance,and then the KD-tree search strategy was introduced to improve the speed of feature matching.Finally,the random sampling consistency algorithm was used for matching purification to eliminate false matching.Experimental results show that compared with the target recognition algorithm based on SURF,the improved recognition algorithm improves the recognition rate of the tank interface from 79.3%to 92.7%,and the average time from 1.504 s to 1.214 s.Thirdly,this thesis introduces the principle and evaluation criteria of stereo matching,analyzes the basic principle of the traditional semi-global matching algorithm,and points out its limitations for parallax calculation in the weak texture and depth discontinuous area of the fuel tank interface.Then,this thesis proposes an improved semi-global stereo matching algorithm,which combines median filtering with Census transform to improve the calculation method of cost,and then builds a cross domain for aggregating matching cost.In parallax calculation and uses sub-pixel fitting to improve the accuracy of parallax calculation,finally get higher quality parallax image of fuel tank interface and the algorithm time was reduced from 5.710 s to 3.127 s.Finally,the internal and external parameters of the binocular camera are determined according to Zhang Zhengyou’s calibration method,and Bouguet algorithm is used for stereo correction of the left and right cameras.Then acquire a parallax image of the corrected tank interface and based on the parallax image and reconstructs the 3D point cloud in the identified interface area.And this thesis uses Hough circle detection to determine the 3D spatial coordinates of the center of the interface circle and segments the interface plane point cloud,and at last uses RANSAC algorithm to fit the point cloud plane,calculates the plane normal vector,and completes the pose estimation of the interface.Experimental verification shows that the spatial positioning accuracy of pose estimation is less than 2mm and the angle positioning accuracy is less than 2°,which meets the needs of the binocular visual positioning system of intelligent refueling robot.