Research On Vision Position Technology Based On Features Template Library

Positioning technology based on computer vision is a new technologydeveloped rapidly in the field of industrial robots in recent years.Because of itsfeaturing high precision,non-contact,real-time analysis and control,continuousoperation,it has been applied in many fields,and provides a new way ofextending intelligent robots smart positioning,and improves the productionefficiency and intelligent degree of traditional industrial robot. How to identifythe target and accurate positioning through machine vision is the purpose of thisthesis. The main works are as follows:First,according to the color characteristics of the experimental environmentand objects, foreground and background separation algorithm based on a colorimage,which can effectively crude extract for identify object,was proposed.Second,the median filtering and image enhancement is used to improve thequality of the image. Region segmentation is required before target recognition,and by comparing several segmentation method,the maximum variance methodfor image segmentation was selected.Then,some methods for target identification was introduced.In this thesis, aspecific identification method was proposed to recognize objects on experiment(including circular, rectangular, cross, etc). Because these objects will occurvariation of translation, rotation, and size in the process of image projection, amethod which establish a training set of feature for template matchingclassification was established to recognize target, these features ensured affineinvariance in the process of the target recognition, and improved the robustnessof the algorithm.In addition,the objects’ geometric center of gravity wascalculated, and an improved sub-pixel center positioning method,which caneffectively improve location accuracy,was used.Finally,target position is further analyzed.Based on one camera, internal andexternal parameters of camera was calculated through the calibration technique，and the known size of the target,the deepness coordinate of the target could becomputed,then other two coordinates.The experiment validated the feasibility of this method.A large number of experimental results prove that the algorithms can berealized quickly and accurately identify and locate the object within the scenewith better usability,stability,and effectively improve the recognition accuracyrate of objection.