Research On Recognition And Positioning Of Stacked Parts Based On Binocular Vision

In the context of the intelligent era,with the development of intelligence,robots equipped with vision systems are equivalent to allowing robots to have "eyes" to promote more intelligent and flexible industrial manufacturing.Industrial sites often need to complete the work of positioning disordered stacks of parts,and because monocular vision is not a good solution to this problem,it is necessary to rely on manual or vibratory mechanisms to place disordered stacks of parts flat,and then use monocular vision for subsequent operations,affecting production efficiency and increasing manufacturing costs.In order to improve the production efficiency of enterprises and promote the further development of automation,this thesis addresses the problem of disordered stacking by conducting a study on the identification and positioning of stacked parts based on binocular vision,with the main research content as follows.(1)Research on binocular calibration and hand-eye calibration.The theoretical analysis of binocular calibration,calibration and correction experiments were carried out on MATLAB,and the reprojection error was 0.21 pixels,indicating a good calibration result.The hand-eye calibration matrix was solved by controlling the end of the manipulator with three random translations and one rotation.The joint manipulator was verified for positioning accuracy using Euclidean distance deviation,with deviations in the range of 0.5 to 1.1 mm and a mean value of 0.78 mm indicating high accuracy.(2)Research on pre-processing and segmentation algorithms for disordered stacked parts.The images are pre-processed according to the actual working conditions of the disordered stacked parts and a hybrid filter is designed,which is shown to have better suppression effect than a single filter.The Canny operator is used to detect the edges of the part maps of different working conditions,and the segmentation algorithm of the colour marker watershed for morphological processing is proposed for the problem of its inability to locate accurately.The results show that this algorithm has a high segmentation rate,a small error rate,a high efficiency and can accurately delineate the ROI region.(3)Research on stereo matching algorithm for disordered stacked parts.The improved SGBM semi-global stereo matching algorithm based on foreground segmentation,"black hole filling" and mis-matching point rejection is used to obtain parallax,and the experiments show that the improved algorithm can better solve the matching problem of weak textures and occluded areas,improve the matching effect and is efficient.(4)Research on the 3D reconstruction and positional acquisition of disordered stacked parts.The ROI region is set by the segmentation result,the edge contour is detected by the edge,the 3D coordinates are calculated by using the parallax,the spatial circle is fitted on MATLAB using the random ellipse method to obtain the bit pose,and the grasping priority is set according to the position height z of the part and the horizontal distance x between the part and the camera.(5)Build the platform and grasping experiment.The spatial coordinate values were obtained through 3 times sampling points 6 times sampling,and the difference was in the range of 0?1mm,with a small fluctuation range,indicating high repeat positioning accuracy.Through6 times sampling points 2 times sampling,calculate the distance between two points in turn compared with the actual,the deviation is less than 1mm,and the average value is 0.3mm,indicating high measurement accuracy.Through different numbers of different stacking state grasping experiments,the results show that the grasping success rate is 96.88%,the grasping time fluctuates between 1.8s?2s,and the experimental results are good.It shows that the system in this paper has high accuracy,good stability and feasibility.