Research On Object Detection On Train Bogies Using Structured Light Scanning

The train bogies inspection is an important measure of railway operation safety.Traditional inspection is completed by rail technician,which is costly and inefficient.With the development of digital image processing technology,the application of machine vision to train bogies inspection becomes a hot research topic in the field of intelligent train fault detection.Compared with 2D machine vision,3D vision can directly obtain the size,the shape and the volume of an object,and overcome the contrast problem and color interference.Therefore,this study focuses on the issues of train bogie inspection and object detection with 3D vision,and the work content is given below:First,an experiment platform for train bogies inspection and object detection was established,which consists of a structured light scanning module,a photoelectric scan starter,a train bogie module and a computer.The structured light scanning module is composed of a linear laser and a monocular camera,which obtains the laser stripe images as the train bogie runs in front of the module.The photoelectric scan starter has two pairs of photoelectric switches and a microcontroller,and the trigger signals from switches are used to start or stop the scanning.The train bogie module is composed of a movable bogie and a control console.The computer controls the system devices,and realizes image acquisition and data visualization.Meanwhile,software codes for 3D measurement of train bogie was designed and implemented.Second,the 3D reconstruction and surface reconstruction of train bogies are implemented using the scanned images.The 3D reconstruction of train bogies is based on the principle of triangulation measurement.The parameters of the camera,the optical plane and the base plane of the structured light scanning module are calibrated.The stripe images of the train bogies were preprocessed by a threshold segmentation method,and then the central pixels of the light stripe are extracted by a refining method and a gray centroid method.The coordinates of the central pixels of laser stripes can be computed with the calibrated parameters in terms of triangular geometry principle,and the 3D point cloud of the train bogie is obtained.A greedy projection triangulation algorithm and a Poisson algorithm are used respectively to reconstruct the surface of 3D point cloud of the train bogie to obtain better visualization.Third,model difference and hierarchical clustering algorithms were used to detect objects on the train bogie.The model difference method is used to segment the point cloud of the object on the train bogie,and using the characteristics of these point cloud,the adaptive hierarchical clustering is proposed to get rid of the noise of the segmented points and classify the object points.Finally,the aspects of degrading the 3D measuring accuracy of structured light scanning are analyzed deliberately.Here,the emphasis is put on the illumination,the bogie structure,which are the dominant factors to cause the measurement errors.A comparison study was made to evaluate the measurement accuracy of the volume of the regular and irregular objects.In this study,an experimental platform for train bogie inspection and object detection was established,to realize the 3D reconstruction and the surface reconstruction of a train bogie and object detection on it.This work will provide an object detection technology with high efficiency and low cost,and promote the development of intelligent detection technology in the area of train routine inspection.