Research And Implementation Of Contour Matching Algorithm For Embedded Vision Sensors

With the development of modern industry,machine vision is more and more widely used in industrial production,and the implementation and optimization of machine vision algorithm in embedded system has become a hot research direction.Contour matching algorithm is a widely used machine vision algorithm,but there are some problems in the practical application of the contour matching algorithm.First,the performance of algorithm is low,and the real-time demand cannot be achieved.Secondly,the application environment is complex,such as the unstable illumination,so the robustness of the algorithm needs to be improved.Third,the application requirements are complex and diverse,and the general applicability of algorithms is challenging.Aiming at the above problems,this thesis designs and implements an embedded vision sensor contour matching algorithm.The algorithm's feature is shape gradient,and optimizes the algorithm according to the characteristics of embedded system,which has the advantages of fast matching speed,high robustness and versatility.The innovations and advantages of the algorithm include the following four: First,based on the image pyramid optimization method,combined with the integral graph and the integral graph acceleration strategy based on the weak classification,a three-level method for quickly obtaining the target candidate pose on the top of the image pyramid is designed.This method can quickly obtain the coarse positioning and coarse angle of the target contour on the top of the image pyramid,which greatly improves the performance of the algorithm.Second,For multi-layer matching process in the image pyramid,when the target candidate points are expanded,a large number of repeated extended candidate points will be generated.This thesis designs a method for obtaining the optimal extended candidate points in the target candidate point neighborhood.So the candidate points can be effectively expanded in the set of interlayer candidate points,and the problem of repeated calculation of the expanded candidate points is solved.Third,for the problem of insufficient computing power of the embedded system,static lookup table and dynamic lookup table are used to optimize the algorithm,which can reduce the amount of floating point operations.Using a table lookup method to record and find sine and cosine results and algorithm runtime state can reduce a large number of floating point operations.Fourth,it provides a configurable interface to the key parameters of the algorithm,ensuring high robustness and high applicability of the algorithm in different application scenarios.In order to verify the effectiveness and performance of the algorithm,this thesis designs and implements a functional test and a performance test for the contour matching algorithm.Functional testing was performed for the functional requirements of chirp detection and the robustness of illumination variations and complex contexts.Performance test is performed to test the time it takes for the algorithm to create the contour template and the time it takes for the contour template to match.The test results show that the algorithm satisfies the requirements of common flaw detection in the above scenarios,and has good robustness and meets the expected functional requirements of the algorithm.In the process of establishing contour template and template matching,the algorithm can make good detection and judgment within the expected time,it proves the effectiveness of the optimization method of the algorithm.