Research On Laser Collimation Measurement Technology Base On Multi-cameras Real-time Stitching

Track irregularity is an important factor restricting the high-speed development of railway.A lot of information indicates that the irregularity of the track is the main cause of the violent vibration of the train,and severe track irregularities can threaten train safety,we must be taken seriously of it.Therefore,railway check equipment is developing towards the direction of high precision detection of track irregularities.Among them,the application of laser collimation technology to orbit measurement is the focus of current research on orbit smoothness measurement.However,current laser railway check equipment's measure module using single camera,which is bulky and low accuracy.In order to solve this problem,this paper proposes a such of measurement about laser collimation measurement system which base on multi-cameras real-time stitching.In this system,four cameras form a matrix array,and using image stitching technology to stitch four cameras' image,then detecting the laser spot position based on the stitched images.Compare with single camera measurement,this measurement has the smaller size and higher precision laser spot location measurement.According to the demand analysis,this paper researches and implements the proposed laser collimation measurement system based on multi-cameras stitching.Firstly,this paper analyzes the practicability of the measurement system after the introduction of multi-cameras,determines the matrix array method as the camera arrangement scheme.In the meantime,select the cameras and lenses involved in this paper,use Zhang Zhengyou's calibration method to correct the camera and lens distortion errors.And,built a prototype of the measurement system.Then,combine measurement system prototype and four cameras arrangement,this paper researched four cameras' image real-time stitching technology,which include image preprocessing,image registration and image fusion.The key parts are image registration and image fusion.In this paper,SIFT is selected as feature extraction algorithm in image registration and improvements are made.We divide the image overlap region into ROI regions,and detect and match features in the ROI regions,improve feature matching accuracy.In feature matching.we use RANSAC algorithm,which effectively eliminates many anomalous matched points.Because the camera position of this measurement sy'stem is fixed.the position between capture screen and cameras is also fixed,so image registration only needs to be done once.Calculating the homography matrix between cameras and store it locally,and follow-up image stitching without computer homography matrix again,which guarantees the work efficiency.After we get the homography matrix and use it to registration images,the Alpha blending algorithm is used for image blending,which makes the image blending more natural without any obvious stitching seam and realizes the four cameras' image stitching effectively.The final experimental results are as follows:single camera gets image in 6fps,single image's distinguishability is 2048×1536;the stitched image's show speed is 2fps,because there are overlapping areas between images,so stitched image size is 3429x2791.Then,using the gray barycenter algorithm to extract the laser facula location coordinate,which can meet 0.1 pixel accuracy.At the same time,we computer this measurement system's pixel equivalent is 0.0683mm/pixel.Besides,this paper introduces the user graphical interactive interface written in C++and using Qt as the development platform,and real-time display the stitched image and coordinate extraction results on the interface.In the end,this paper does the stitch experiment and measurement experiment.On the basis of experiment result,this measurement system can stitch four cameras' video image in real-time.And system's measurement accuracy is 0.0841mm.