Research On Location And Posture Recognition System For Scattered Bars Based On Random Sample Consensus

In industrial automation production,the use of robots to grab parts is a hot spot in intelligent manufacturing research.The accurate positioning of industrial parts is one of the key technologies for robotic grabbing,and the positioning of scattered and stacked parts in the accurate positioning of industrial parts has always been a difficulty.In this paper,the scattered bar-like parts are taken as the research object,and the collection of bar point cloud data and the pose recognition algorithm based on point cloud processing are studied.The specific work includes the following aspects:(1)A method for obtaining point cloud of scattered stacking bars based on line structured light is analyzed.A simplified mathematical model for three-dimensional vision measurement of linear structured light is established.The camera calibration method,the structured light plane calibration method and the line structure light stripe centerline extraction method are discussed.Point clouds scattered with bars were captured.(2)The preprocessing algorithm for scattered point cloud of bar is studied.According to the characteristics of the cumulative distribution of the scattered point cloud along the z-axis,the z-directed pass filter is used to preserve the target point cloud within the filtering range;the statistical filter is used to delete the isolated outliers in the bar point cloud;voxel grid filter is used to simplify the point clouds,whichimproves the efficiency of the subsequent algorithm for processing the point cloud.(3)Random sampling consistency algorithm and improved nearest point iteration algorithm are used to study the pose recognition method of scattered stacked bars.For the bar point cloud with cylindrical features,some points are randomly selected from the input point cloud to calculate the initial value of the input cylinder model,andthe distance threshold t is used to select the points satisfying the model,and the number of points S in the model is counted.This process is iterated k times,retaining the largest cylinder model,and re-estimating the model as the best cylinder model for output,and then separating the bar point cloud from the scattered point cloud.The nearest point iteration algorithm with SAC-IA algorithm is used to segment the bar point cloud and get its position information,and finallythe position and posture of the bar are obtained.(4)The hardware system of pose recognition for scattered stacked bars is built,and the experiment of position and posture recognition was carried out.The software of pose recognition system is compiled on the platform of windows,and the pose recognition was carried out on the scattered stacking bar in the space of 200mm*200mm*150mm.The rotation angle of a single bar is identified by means of an optical rotating platform,and the error of the recognition angle is ± 0.5 degrees.The position and attitude of the bar are recognized as a whole,and the recognition angle error is ± 0.3 degrees,and the maximum erroroftherecognition position in the z-axis direction is 0.733 mm.The experimental results show that the proposed pose recognition method based on random sampling consistency can accurately identify the position and posture of scattered stacked bars,and can be widely used in the position and posture recognition of rod parts and robot guided grabbing.