Font Size: a A A

Visual Detection And Identification Of Industrial Robots Under Motion Background

Posted on:2018-12-12Degree:MasterType:Thesis
Country:ChinaCandidate:C Q ZhangFull Text:PDF
GTID:2428330572452620Subject:Mechanical engineering
Abstract/Summary:PDF Full Text Request
Machine vision is one of the important application technologies in digital image processing in industry,which is also an important technical foundation in the field of artificial intelligence.The goal of machine vision is to replace human vision to complete industrial work more efficiently and accurately,realizing industrial automation and reduce the labors.Under the developing background,automation requirement for the industrial site is becoming more and more strict,industrial robot detection and identification of the target has become one of the hot and difficult research in the field of industrial 4.0 advanced manufacturing.The key technologies include:1)Under the motion background,the first step is to obtain the initial position of the workpiece crawled and extract the feature of the detecting object,which can separate the target from the complex motion background and obtain the real-time position of the moving target;2)Identify the detecting target pair on the basis of classification and realize the semantic segmentation of artifacts under the motion background.This thesis regards the target detection and recognition method of industrial robots as the research object,aiming to find a visual inspection algorithm for industrial robots under the motion background.This method can realize reliable visual tracking and the semantic segmentation of artifacts.The main work of this thesis include:1.The industrial robot platform established is applied to visual detecting and identification tasks.There are five sorts of the image database for mechanical tools and parts,such as bearing,screwdriver,gear,pliers,wrench etc.database is mainly including two types of images,which are single image single target single category and single image multi-target multi-category.2.Aiming to the disadvantages of the traditional machine vision detection method that manually assign the initial position of the target in the first frame,this thesis put forward a method of target detection based on the background difference that can automatically obtain the initial position of the target in the first frame.Combined with the Spatio-Temporal Context target tracking method,the industrial robot.Can track the workpiece target on the conveyor reliably.3.The depth learning method is applied to the classificating identification of mechanical tools and parts targets by industrial robots under the motion background.The whole-convolution neural network model is trained by 1938 workpiece target images,which leads to the semantic identified classification under the motion background.
Keywords/Search Tags:industrial robot, target tracking, classified identification, convolution neural network, fully convolution network
PDF Full Text Request
Related items