The Research And Application Of Image Feature Extraction And Matching Algorithm Based On Geometric Congstraints

Image features are widely used in image matching applications in view of image features have scale,rotation,and translation invariance.The algorithms involved in image feature extraction are SIFT and SURF.Image feature matching is an important link in industrial machine vision processing.However,due to the harsh industrial environment in the field of machine vision,noise is easily caused,artifacts such as grits are attached to the workpiece,and the image disturbance in the background makes the image features.In the process of feature point detection,the extraction algorithm has spots which interfere with the image,resulting in the inaccurate matching of the image.At the same time,the image feature extraction algorithm has two disadvantages in practical applications:one is that the image feature extraction algorithm does not distinguish the background and foreground of the image,which makes the detected feature point appear in the background image,which will cause the incorrect matching of the image feature points in the foreground.Another is that the common parts have rich geometric texture features in the industrial scene,and the feature extraction algorithm can not use these accurately.The effective information extracts and matches the image features,and even the Gaussian process in the algorithm will cause the edges of the image to be blurred.This paper starts with the two major problems,complements the feature points of the original algorithm on the edge of the target workpiece further,replenish the feature points of the SURF algorithm at the edge of the target workpiece and expands further the feature extraction and image matching algorithm for the research and application.The main research content of this article is as follows:Firstly,in order to eliminate interference of the background image in industrial application field;first of all,this paper analyzes the difference between the background and the foreground in the image,and notices that the target workpiece in the foreground has rich geometric characteristics,but there are no obvious specific features in the background.Secondly,the geometrical characteristics of the length,perimeter and area of the edge curve in the target artifact and the background binary image are studied and analyzed.Finally,according to the difference of the geometric characteristics of the workpiece,two filtering interference curve treatment schemes are proposed:one with thresholds and another one without thresholds.Each of two processing schemes includes three filter rules based on the three geometric characteristics of the curve.At the same time,this paper analyzes and compares the filtering effects of each filtering rule.The experimental results show that both schemes can effectively extract the shape curves of the target at different scales.Secondly,in order to take full advantage of the rich geometry of the workpiece;this paper proposes two geometric optimization processing schemes:one is to perform geometric optimization processing from the perspective of different scales of the image in the algorithm,and the other is to perform geometric optimization processing from Hessian matrix traces of different scales in the algorithm.Through the combination of filter interference curve rules,two sets of optimization schemes are analyzed in turn.It can be seen that the two sets of programs can obtain the target shape curve.This paper also improves the SURF algorithm to match the line features of the image better,which replace the point features of the image.First,the Hessian matrix trace scheme is used to optimize the image processing.Secondly,the paper uses the filter rule to filter the background interference curves of Hessian matrix traces at different scales to obtain the shape curve of the target image.At last,the detection condition of the feature points is modified in the algorithm,which makes the algorithm display more feature points in the edge area of the image.At the same time,the detected feature points are filtered to obtain the line features located at the edge of the workpiece in images of different scales.Then using the description operator of the line feature of the SURF algorithm to match the image,and combining the RANSAC algorithm to eliminate the feature points that are incorrectly matched.Thirdly,in order to verify the effectiveness of the improved algorithm,this paper tests it from two aspects:objective evaluation and subjective evaluation.In the objective evaluation,the algorithm's processing results are analyzed by using four objective indicators:average point error distance of feature points,repetition rate of feature points,feature point matching accuracy and the PR curve.The experimental results show that the number of effective feature points of the improved algorithm increases,which improves the matching of image line features.In the subjective evaluation,this paper uses the scene survey method and applies the improved algorithm to carry on the preliminary test to the image which the actual industrial robot collects:Firstly,the paper carries on the image feature matching processing to the acquisition single target image.Secondly the paper carries on the characteristic point to the multi-objects detection process.Through experiments,it can be known that the improved algorithm can better detect the edge line characteristics of the target workpiece when processing the image collected in the actual factory environment,which could provide a theoretical basis for the further application of industrial robots.