Research Of Visual Detection Algorithm On Engine Cylinder Parts Manufacturing

Engine cylinder head is one of the key parts in the engine.Unqualified size of the holes on engine gives rise to the bad valve seal,which reduces the energy efficiency and aggravates environmental pollution because a large amount of energy with high temperature gas is discharged from the defective valve.So it is imperative to accurately measure the size of the holes on engine.Combining the demands of military project in Chong Qing,this paper realized an automated and highly precise detection for the parts of engine cylinder head with computer vision technology,and researched the related visual algorithms.Based on the research of engine cylinder head detection,firstly,this paper summarizes the research status of the product detection,feature matching and Convolutional Network.Then the difficulties of measuring the size of the holes on engine cylinder head is expounded and analyzed.Finally,this paper introduces the theories of the optimization model,the feature matching algorithm and convolutional network.For the problem of local texture difference in round areas caused by the angle of camera and brightness of light,this paper proposes an optimization model based on gray threshold search.Firstly,the objective function and constraint condition of the optimization problems are constructed based on subspaces obtained by the gray threshold search,and the optimization problem is solved to obtain the circular integral region.Then the edges of the measured circle are corrected by the completeness and smoothness of contour to improve the measurement accuracy.The experimental results show that the optimization model can effectively reduce the influence factors of light and have high precision.For the problem of the high precision measurement,this paper presents a novel detection algorithm based on feature point matching.In order to improve measurement accuracy,it is necessary to reduce the distance from the camera to parts surface and increase resolution,which will decrease overall difference on gray level between circular and non-circular region and greatly affect the measurement results of the optimal model based on gray level threshold search.The feature point matching depends on the local texture feature,and the texture for the same type of engine cylinder head is similar under the same environment.So it can realize high precision measurement with high resolution.Firstly,this paper uses the edge feature to estimate the potential regions of the matching points.The improved Haar-like feature is used to further identify the most similar matching points to increase the accuracy.Finally,the effects of light and texture difference are reduced by improving the image similarity or coding algorithm and enlarging the template library.The experimental results show that the point feature matching algorithm based on the edge and improved Haar texture features is robust to illumination changes,and the accuracy errors of the detection radius is 0-1 pixelFor the selection of effective features,this paper proposes a new convolution matching network based on FCN and RCF network.The edge feature and improved Haar texture feature are extracted by subjective consciousness based on the problem analysis.So there are certain limitations for the effectiveness of features.Through the convolution neural network,the stronger dependence between a pixel and other pixels in the neighborhood is automatically learned.In this paper,the convolutional matching network is composed of two channel blocks with different convolution kernel scales.The rich dependencies between pixels are obtained by extracting convolution characteristics of each layer from each channel.The significant Euclidean space and color space are used to construct the sample similarity label,and the normalized mean square error is used to train the network.The experimental results show that the convolution matching network can receive stronger pixel dependences,more accurate feature points and higher precise results.For the requirements of the research and project,a machine vision measurement system based on the engine cylinder head parts is designed and implemented.This paper analyzes the measured results of the upper side of engine cylinder head parts,which satisfies the precision requirement of project measurement.