Research On Binocular Vision System Of Automotive Refueling Robot

The auto refueling robot can make up for the deficiency of the existing refueling mode by automating the refueling,so it is of research value to configure the auto refueling robot and realize the intelligent refueling in the auto gas station.In the present research,there are many researches on the fueling mechanism,the fueling interface,the robot structure and the motion characteristics of the refueling robot.There are relatively few researches on the identification and positioning technology of automotive refueling robot,especially the research on the application of binocular vision to the system of automotive refueling robot.Therefore,this paper designs a set of binocular vision system for the identification and positioning of the automotive refueling robot.The main contents are as follows:(1)Construction and calibration of binocular vision system.Firstly,the hardware is selected and the function of each module of the system is designed.Then the imaging model and calibration method of monocular camera are studied,and the measurement principle,calibration method and stereo correction method of binocular vision are studied based on this.Finally,the calibration experiment is completed and the experimental results are analyzed.(2)An algorithm for target recognition and location of vehicle fuel tank cover in unstructured environment is designed.Firstly,a candidate box extraction method based on multi-scale bottom-hat transformation algorithm and improved maximum entropy threshold algorithm is proposed.The image is collected and the sample data set is made by labeling.The HOG feature is extracted from the sample set and a training method of support vector machine classifier is designed.Finally,the experimental results show that the efficiency and target detection rate of the proposed algorithm are above.(3)Aiming at the problem of image acquisition of fuel tank cover in application scene,an edge detection algorithm suitable for vehicle fuel tank cover was proposed.Firstly,a variety of edge detection algorithms are studied and tested.Aiming at the problems of edge noise,edge missing,edge dislocation and edge redundancy,an improved Canny edge detection algorithm based on pseudo spheres is proposed.Experiments show that this method has good effect in edge location and noise suppression.On the problem of removing the target edge,an edge extraction algorithm based on morphology is designed.This method can obtain the complete edge of the tank cover plate but the positioning accuracy is poor.The two edge detection methods are combined to obtain a complete edge of automobile fuel tank cover with high positioning accuracy.(4)To solve the problem of three dimensional positioning of automobile fuel tank cover,a method of three dimensional positioning of automobile fuel tank cover is designed.Firstly,the error sources of 3D contour edge reconstruction using binocular vision system are analyzed.Then,a stereo matching method for the edge profile of the tank cover plate is proposed.By selecting the edge points,the stereo matching error is reduced to improve the reconstruction accuracy.Finally,the spatial position and grasping direction were obtained by plane fitting of the spatial points,which were verified and analyzed by experiments.(5)In order to detect the elliptical features projected by the vehicle fuel tank cap in the image,the existing arc combination algorithm was improved.The ellipse fitting method based on algebraic distance and geometric distance is studied,and a two-step ellipse fitting method is proposed to screen effective arcs and effective arc combinations,and finally the ellipse parameters of fuel tank cap are obtained.Finally,the feasibility of the proposed method is verified by experiments.(6)In order to solve the pose estimation problem of vehicle fuel tank screw cap,a new pose estimation method of space circle based on projection curve parameters was proposed.The validity and accuracy of the proposed algorithm are verified by experiments,and compared with other measurement methods,the results show that the proposed algorithm has better stability and accuracy.