Research On Dual-arm Robot Intelligent Sorting System Based On Machine Vision

With the rapid development of modern manufacture,robots are playing a more and more important role in sorting system.Robots can finish heavy,repetitive and highaccuracy sorting work more efficiently compared with human beings.And their actual costs are lower.At the same time,with the development of machine vision technology,using the machine vision technology to lead robots to finish sorting work is becoming a trend in intelligent sorting.Therefore,using machine vision algorithm to recognize the object and to achieve fast sorting is of great practical value.This thesis focuses on the intelligent sorting technology based on machine vision,researches recognition technology of objects with small differences,and aims at combining advantages of Kinect and Baxter dual-arm robot to develop the intelligent sorting system with perfect interaction,high security,low cost and great versatility.First of all,as far as the problems of image pre-process based on Kinect is concerned,this thesis uses mid-value method to repair the missing depth information,and uses bilateral filter to process the repaired depth information according to the reasons of missing depth information and the actual situation.Being aimed at RGB image with low resolution and the effect of illumination,local threshold based on connected component labeling is proposed to segment objects from the RGB image,which has been preprocessed by bilinear interpolation to improve image resolution.And contours of objects are extracted from the binary image obtained from adaptive threshold.At the same time,in order to improve the calculation speed of feature extraction,polygonal approximating algorithm is employed to reduce the number of points in the contours.Secondly,object recognition based on geometric features is studied.In order to recognize objects with small differences,this thesis creates a sample set,which extracts Fourier descriptor,invariant moments feature and basic geometric features as the main features,and researches every feature’s ability to express the object.As the result,feature fusion method based on fuzzy C-means(FCM)is proposed.Using this sample set,Supper vector machine(SVM)is employed to train classifier,which can accurately predict classification of unknown objects.Finally,using Kinect and Baxter dual-arm robot as the hardware foundation,an intelligent sorting platform is built to research the calibration between Kinect and Baxter robot.Error modeling which is used to compensate errors of points is also built to establish relations between errors and coordinates in the workspace.Sorting experiment based on the sorting platform has been done,and the result shows that the reliability and stability of the sorting system are able to meet the requirement of sorting,and achieve the desired goal.