Image Processing Method In Laser-scanning Photogrammetry Of Large-scale Components

In the process of manufacture and assembling large aeronautical components,it is necessary to ensure their accuracy,efficiency and reliability through measurement.At present,photogrammetric system represents large aeronautical components as point clouds.By comparing the difference between point clouds and theoretical models,the geometric accuracy of components can be verified.Because of the high straightness and stability,laser stripe has been widely used in vision inspection systems to enhance features of target images.As the key technology of photogrammetric,image processing directly affects the robustness and accuracy of the system.In the past,a strict inspection scene was established to avoid difficulties in the image processing,which is expensive a nd not applicable to various scenarios.However,this is not suitable for measuring a variety of large-scale aeronautical parts.Due to the inability to construct a widely applicable photogrammetric environment,the image contains various complex background noise,which makes image processing very difficult.With the development of intelligent technology of image processing,it is feasible to extract the laser stripe in a specific industrial environment.However,due to the interference of light spot,background object,dust,light source and dynamic scene,it is impossible to extract the target accurately.In this paper,a binocular stereo vision measurement method based on auxiliary laser scanning is used to measure point clouds of large aeronautical components.The methods of extracting laser stripes,including localization,segmentation,denoising and compensation,have been studied as follows:(1)Because the photogrammetric image of large aeronautical components contains large pixel size and complex background noises,it is very difficult and inefficient to separate laser stripes from the image directly.Therefore,it is necessary to locate the region of laser stripes in the image in order to reduce the processing range and remove most of the redundant background noises.Then,a location method of laser stripes based on multi-feature maps has been proposed.The proposed method is validated in two groups of experimental images.The experimental results show that,the proposed method takes 2.875 seconds/fra me and can deal with all experimental images,while the location method based on Kalman filter takes 3.116 seconds/frame and cannot locate targets in the second experiment.Thus,the proposed location method of laser stripe can deal with the abrupt change of position and size of the laser stripe,with higher efficiency,wider application range and better robustness.(2)Due to the interference of background noises around the laser stripe,calculating the center line of the laser stripe directly will lead to great errors.In order to segment the laser stripes and background noise accurately,a fast and reliable segmentation scheme is proposed based on threshold segmentation method: selecting the best binary images of laser stripe.Furthermore,a fast threshold segmentation method based on supervised learning is proposed.The proposed method is validated in three kinds of experimental images.MSE,SSIM and IoU are used for evaluation and the proposed method is compared with image segmentation methods of GLGM,Grabcut and FCN.The experimental results show that,the proposed segmentation method of laser stripe has the best accuracy and stability.(3)In some cases,there are some noises covering the laser stripe,resulting in the loss of key measurement information.In order to recover the lost information,the method of denoising and compensation is studied.The noise covering the laser stripe is classified into linear noise and block noise based on the morphology.Then,the methods,including distinguishing linear noise from laser stripe,compensating linear noise based on the width of laser stripe and filtering block noise based on morphological skeleton,are proposed.The experimental results show that,the proposed methods can effectively filter covering noises and compensate defective laser stripe.(4)A binocular stereo vision measurement system based on auxiliary laser scanning is built.Three experimental images with different laser stripes,different scenes and complex background noises are obtain.Firstly,the location and segmentation methods of laser stripe proposed in this paper are verified and compared with the classical methods.Then,methods of denoising and compensation are verified.Finally,the sub-pixel centerlines of laser stripes are calculated and the point clouds of the aeronautical parts are further obtained.