Research On Recognition And Location Of Stacked Workpiece Based On Machine Vision

At present,China has become the second largest economy in the world,but the independent innovation ability of the manufacturing industry is slightly insufficient,and key core technologies are restricted by others,therefore,the country has clearly put forward the development strategy of "Made in China 2025" to build an internationally competitive manufacturing industry.As an important part of the manufacturing industry,the automation level of workpiece sorting is still low in the face of complex working conditions.Aiming at the sorting problem of stacked workpieces in unstructured environment,this theses proposed a method of identifying and locating stacked workpieces based on machine vision,and verifies the effectiveness of the proposed method by guiding industrial robots to grab the workpieces in the field of view through a monocular structured light vision system.The main research work is as follows:Firstly,the establishment of the camera model and the transformation of relevant coordinate systems were presented,and the production process of the project workpiece dataset was introduced.The particularity of the workpiece image in the subject was given.Aiming at the problems of image contrast,the contrast experiment was verified by histogram equalization and adaptive histogram equalization with limited contrast,and this theses has introduced the related information of dataset making software,and completed the stacked workpiece dataset.Secondly,aiming at the problem of traditional object detection algorithm to identify stacked workpieces,a stacked workpiece detection method based on improved YOLO v3 algorithm has proposed.Inception structure has introduced to enhance the feature extraction ability of the feature detection network,and enhanced feature pyramid structure has introduced to improve the multi-scale feature fusion ability of the model.The anchor frame of the workpiece has re-determined by using the intersection ratio distance of K-means clustering fusion,and the effectiveness of the method for the recognition of stacked workpieces has verified by experiments.Thirdly,aiming at the problems of point cloud data registration of objects with few texture features,a point cloud registration method based on point neighborhood condition constraints has proposed.The feature points was extracted by using the internal morphological descriptor and the change of the normal vector of the point cloud.The efficiency of feature extraction of the internal morphological descriptor has improved by using the double constraint method of point neighborhood information weighting and adjacent point searching.The histogram feature description ability has enhanced by using the weighted distance constraint of the adjacent points,and then the initial registration and accurate registration of the object point cloud data were completed.Experiments have verified the effectiveness of the proposed method in improving the registration accuracy and efficiency of the workpiece point cloud with less texture features.Finally,a human-machine interaction system for identifying and locating stacked workpieces has built,and the software and hardware design and debugging of the system has completed.The effectiveness of the method proposed in this theses for identifying and locating stacked workpieces in unstructured environment has verified through experiments.