Binocular Vision Ranging Method Based On Super-pixel Grid Motion Statistical Feature Matching

Laser marking is widely used in electronic devices,precision instruments,hardware tools,materials and appliances,etc.With the rise of private customization and other industries,the types of marked workpieces become more diversified.Laser marking can realize non-contact processing by virtue of its superior characteristics,which is not affected by tool bit wear,with high processing efficiency,environmental friendliness and pollution-free.However,the marking quality is directly affected by the focusing accuracy of laser on the workpiece surface,and only one kind of workpiece can be processed on one assembly line,and there are certain requirements on the placement position of the workpiece.Therefore,in order to improve the machining quality,accurate identification and position detection of the workpiece and accurate measurement of the distance between the laser exit position and the surface of the marked workpiece become the key points.Aiming at the problem that the existing distance measuring methods can’t meet the marking requirements when the workpiece identification and position detection can’t be carried out at the same time,and the workpiece has too many weak textures and homogeneous areas,this paper proposes a binocular vision distance measuring method based on super-pixel grid motion statistical feature matching.The specific research contents are as follows:Firstly,based on the binocular vision theory and Zhang Zhengyou’s calibration method,this paper calibrates the binocular camera with the consideration of tangential distortion,and obtains the internal and external parameters and distortion parameters of the camera.Secondly,binocular image acquisition is carried out on the workpiece to be measured,and preprocessing such as bilateral filtering denoising,epipolar correction based on Bouguet algorithm,distortion correction based on camera distortion model and so on are carried out to obtain better image information.Thirdly,SOLO instance segmentation algorithm is used to detect the position of the workpiece to be tested and identify the type of the workpiece,which solves the problem that the existing laser marking field can not detect the position and identify the type of the workpiece at the same time.The identification and segmentation accuracy of marked workpieces can reach above 0.99,and the identification speed can reach about 17 ms.Finally,after super-pixel segmentation of the binocular image,the feature points matching ranging algorithm based on super-pixel grid motion statistics is used to match the feature points of the left and right images,and the depth information of the marked workpiece is obtained based on the contrast value of the matched feature points,which solves the problem that the existing ranging methods have low accuracy and cannot meet the marking requirements when the workpiece has weak texture and too many homogeneous areas.In this paper,the method is used to measure the distance between the workpiece to be measured and the initial laser beam position,and the absolute error of the measurement result is within 5 mm.The comparison before and after the improvement of the ranging method for marking workpieces with weak texture and too many homogeneous areas proves that the method proposed in this paper can solve the existing ranging problem of laser marking machine.By comparing the laser exit position of the laser marking machine with the marking clarity of the workpiece to be measured at different distances,the experimental research of this method is carried out within the marking distance range of the laser marking machine,which proves the practical feasibility of the method.