Detection System For Log Stack Diameter Classes Based On Binocular Vision

The detection of log stack diameter classes is an important link in the wood transportation industry.For a long time,the method of using inspection manual ruler results in high intensity of labor and low efficiency.With the development of computer vision technology,the 3D non-contact measurement technology based on machine vision greatly improves the efficiency of the measuring ruler.However,in terms of the detection of log stack diameter classes,the recognition rate,accuracy and efficiency are relatively low because of factors such as the shadow of the pile of logs caused by concave and convex ends and log boundary adhesion.Therefore,this dissertation discussed the key problems of binocular calibration,image segmentation,stereo matching,and the technology of log stack diameter classes recognition.The main work of this dissertation is as follows:First,a three-dimensional measurement model and a lens distortion model based on binocular camera are established.The influenced of the log stack diameter classes recognition rate of binocular camera calibration and distortion correction is analyzed.A modified algorithm for binocular camera calibration based on the distortion model is proposed.The method introduces the distortion factor of three-phase into the distortion model,makes full use of the visual functions of open source to detect the calibration templates precisely,and improves the accuracy of binocular camera recognition rate.With the calibrated and corrected binocular camera,the method will collect the distortionless log stack images.Aiming at the problem of log stack shadow and boundary adhesion in natural environment,an algorithm of image segmentation based on color space transformation and fuzzy theory is proposed.The image data are converted into fuzzy space from the real space.In the fuzzy space,the image fuzzy enhancement is realized by improving the fuzzy membership function and multi-iteration.Meanwhile,fuzzy bandwidth,fuzzy threshold and other related parameters should be calculated.And then,the fuzzy image data are classified and re-identified using fuzzy rules.At last,the fuzzy image data are defuzzified to real space.An log stack edge image comes out.Experiment proves that the method' is more effective than the traditional image segmentation algorithm.Finally,a three-dimensional reconstruction model of the log stack contour is established.The stereo matching optimization algorithm based on feature point detection and epipolar constraint is adopted.The method which extracts the feature points of log contour,reduces the feature points matching search space by combining the epipolar constraint,uniqueness constraint and sequential constraint,and completes stereo matching fast with the strategy of the absolute minimum difference of the pixel value of the matching point.The log three-dimensional coordinates are calculated by using the binocular parallax of stereo matching.The measure of log stack diameter classes and area is completed by using ellipse fitting algorithm based on least square method.The experimental results show that the log stack diameter based on binocular vision measurement meet the relevant national standards and requirements.Compared with the culler measurements,the method which not only has an error within 5mm,but also greatly improves the detection speed and efficiency,has high practical value.