Research On Automatic Detection System Based On Machine Vision For Automobile Insurance Box

The quality of automobile insurance box plays an important role in the safety of automobile circuit.Nowadays,method of detecting defect of automobile insurance box artificially is low efficiency and high cost,which is urging automatic detection system for automobile insurance box to achieve rapidly and accurately screening of substandard products,improving production efficiency and reducing security risks.This paper studies the key of automatic detection system of automobile insurance box.Firstly,on the basis of the actual demand of automatic detection of automobile safety box in production process,studying and designing a vision based detection module and an automatic sorting module based on the existing production model.In the vision based detection module,carrying out designing overall scheme and hardware,which includes placement tool's designing,theoretical analysis and selection of camera,lens and light source,as well as the design of testing platform;In the automatic sorting module,carrying out designing sorting and conveying manipulator,double channel conveyor belt as well as overall work flow of the detection system.Then,,aiming at the problem of a large amount of computation in the traditional matching algorithm for matching the work piece images,especially matching accuracy can not meet engineering practical requirements.On the basis of analyzing different matching algorithms,proposing feature points extracting,description and wrong matching points filtering based on SURF,SIFT and RANSAC,the experimental results show that the matching accuracy is improved.Then,aiming at the problem of the accuracy of rotation correction can not meet engineering practical requirements when the global region image is matching,proposing a diagonal region selection strategy to match the correction through analyzing the relationship between the length of matching points and the accuracy of rotation correction.Compared with using the global region directly,the precision of matching is improved and the computation is reduced.Then,aiming at the problem of error matching and the matching accuracy is not high when global matching,according to the relationship between the coordinates of the diagonal region matching points,proposing the strategy of selecting prepared matching points based on coordinates similarity for getting the best matching point pairs in each diagonal region.As the experiment shows,this method improves the matching precision and effect.Secondly,aiming at the problem of the matching precision of the whole work piece can not meetengineering practical requirements,as the smaller the matching area,the higher the matching precision is,using region segmentation to finely matching each sub region to improve matching accuracy based on global matching.Aiming at low matching efficiency of feature points extracting,description and wrong matching points filtering based on SURF,SIFT and RANSAC,studying feature points description based on BRIEF according to characteristics of different feature description algorithms.As the experiment shows,compared with the SURF algorithm,the BRIEF algorithm improves the real-time performance by 33.7%;compared with the BRISK algorithm,the BRIEF algorithm improves by 15.9%,which improves matching efficiency when ensuring matching accuracy.Then,aiming at the problem of the strategy of selecting prepared matching points is low matching efficiency and robustness,studying improved selecting prepared matching points based on coordinates similarity,which reducing the amount of prepared matching points,improving matching efficiency and robustness.Furthermore,studying fine matching strategy based on unit pixel path searching.As the experiment shows,the matching precision of this method meets engineering practical requirements.Finally,aiming at the limitation of traditional defect detection algorithm,designing a special detection algorithm according to the types and characteristics of the actual defects.For big defects in structural differences,proposing a defect detection algorithm based on morphological open operation;For the defects of extra glue and deserved glue,proposing a defect detection consisting of coarse filter based on mean filter and block size as well as Fine screening based on transverse longitudinal filtering.At last,getting the final defect detection effect by combining the effects of two kinds of defect detection algorithms.As the experiment shows,defect detection accuracy meets engineering practical requirements.