Research On Image Matching Algorithm Of Train Inspection Based On Contour Reconstruction

In recent years,trouble of moving freight car detection system has been gradually promoted to apply to most railway lines,which realizes man-machine detection and avoiding limitations of manual train detection.However,due to various shapes of components and tough external environment,the images of TFDS are of complex background,and the locating of the object region is instable.This paper,which mainly focus on the brake shoe and release valve bonnet,deeply researches matching based on assistance of contour reconstruction and feature description of contour reconstruction.A dynamic staged image matching method based on matching path re-planned by contour reconstruction is proposed.Re-planning the ergodic path by prior experience and reconstructing contours of the objects which have typical geometrical characteristics,then describing the features of object’s shape with Hu invariant moments of contours,meanwhile,coupled with high-low double thresholds,on which whether the second stage matching starts depend.Experiments on brake shoes demonstrate that this method,which ensures the accuracy of brake shoe region location while improving the efficiency,can satisfy the requirements of rapidly accurately locating brake shoe regions.An image matching algorithm based on contour reconstructed by oriented line segment and pairwise geometrical histogram statistic is proposed.Firstly,reconstructing contours by oriented line segments which composed of the sample points based on triangular evaluation element of local curvature,then describing the whole feature of object shape with oriented relative angles and normalized equivalent distances between pairwise oriented line segments,finally measuring the similarity between objects by distance of histogram.The results of the comprehensive experiment on release valve bonnet indicate that the matching algorithm which is well robust against rotation,translation and scaling,satisfies the accurate and real-time requirement of release valve bonnet region locating.Combined with related measures,for instance,Hu invariant moments of contours,radian step,dynamic staged image matching,dynamic sampling based on triangular evaluation element of local curvature and matching by histogram statistic,this paper successfully achieves the rapidly locating of brake shoe and release valve bonnet region.In addition,the matching method of brake shoe can apply to other trains stopped by brake shoes,the matching method of release valve bonnet also can be promoted to apply to the detection of plate-like parts with irregular shapes.