Research On High-precision Three-dimensional Information Detection Technology With Multi-vision Line Structured Light

The road is the most important infrastructure of a city.In recent years,the continuous development and utilization of underground space not only has caused pavement damage,but also led to road collapse,which affect urban traffic.According to the statistics Beijing Municipal Commission of Transport,80%of urban road collapse accidents in Beijing are caused by the failure of underground pipelines which lead to roadbed diseases.Therefore,China University of Mining and Technology(Beijing)cooperate with Beijing Gong Lian Jie Da Highway maintenance Engineering Co.,Ltd.to undertake the key project of Beijing Municipal Commission of Transport that is“Research and application of rapid detection technology for roadbed diseases of urban road in Beijing”.On the basis of this project,this thesis studies on three-dimensional(3D)information detection technology for pavement damages and underground pipeline self-damages.The existing information extraction technologies can't meet the needs of high-accuracy and quantified detections for pavement and underground drainage pipeline.Some detecting methods and their disadvantages are mentioned as follows:(1)Video frequency method is the main method for damage detection.The quantified information of damages or diseases can't be obtained by artificial interpretations.This method is inefficient,inaccurate and easily affected by the environment.(2)Laser radar and multiple line structure lights are mostly used for 3D pavement detections.But laser radar is high-cost and inaccurate for lane-detection and small cracks on pavement;multiple line structure lights cannot achieve high-precision measurement,because its measurement accuracy is affected by exposure time,illumination condition and pavement material.(3)There is not an effective 3D detection method for underground drainage pipeline yet due to constrains of space and environment.In this thesis,3D information detection technology based on line structure light for pavement and underground drainage pipeline is researched to solve the above problems.The research can provide technique support for high-accuracy pavement and underground drainage pipeline detections and also provide precise data for accurate quantified evaluations of pavement damages and underground pipeline's self-damages.The main research work and innovation are as follows:(1)It is difficult to satisfy large-scale and high-precision measurement at the same time with the existing 3D measurement technologies of pavement,and there isn't effective methods for 3D detection of underground drainage pipeline.Therefore,a novel method of large-scale and high-precision measurement with multi-vision line structure light sensor is proposed.In measurement model design,multi-vision line structure light measurement models for drainage pipeline and pavement are respectively constructed to replace the single and multiple line structured light measurement model.According to this model,a multi-vision line structured light measurement method based on image matching is proposed.All the light stripe images are firstly stitched into the main camera's imaging coordinate system.Then,the extraction centers of complete stripe image are transformed with main camera's perspective transformation model into 3D coordinates.In study of image stitching method,the characteristics of light stripe image are firstly studied with image entropy,local contrast,and quality of light stripe center.Secondly,after analyzing the light stripe image features and making registration experiment with the classical methods,it is proved that traditional image stitching methods applied to light stripe images is infeasibility.Finally,considering that the homography relationship of adjacent cameras'images is invariable,a multi-vision light stripe image stitching method based on perspective transformation model for image stitching(PTMIS)is studied.The PTMIS can be pre-calibrated through image registration with the known feature corner points extracted from the calibration images in overlapped field of view.The experimental results of high-precision rack with the proposed and the other three methods show that the proposed method's measurement accuracy is higher if measuring range is same,and its measurement size is wider if achieving the same measuring accuracy.The results of contrast measurement experiment of cement pavement model show that the proposed is more accurate.The average texture depth of the reconstructed model is only 8.3%away from that of real model,which is smaller than other methods.(2)Exposure time is one of the main factors that affect the accuracy of line structured light measurement.Under different illumination conditions,different exposure times are needed to ensure the measurement accuracy of pavement and drainage pipeline.Therefore,in order to obtain the optimal exposure times of different illumination condition,a optimization model of camera's exposure time based on line structured light stripe reliability evaluation is constructed.Based on the introduction of light stripe generation,imaging principle,and the analysis of confidence evaluation's basis,the existing light stripe confidence evaluation model is improved into a novel Gaussian confidence evaluation model by combination with the characteristics that light stripe section energy obey Gaussian distribution.An efficient and precision light stripe centers extraction method for light stripe sample images which specially applied to the optimization of exposure time is studied by the maximum correlation between sample images and the gradient distribution map of LSROI with lowest exposure time.The center extraction results of sample images are evaluated by Gaussian reliability evaluation model.Then the influence mechanism of exposure time on the evaluation energy and the evaluation gray level are analyzed respectively.The cameras'optimal exposure time is finally obtained through analysis of variation models of reliability evaluation results with exposure time.The curve of optimal exposure time for measurement cement pavement in the range of20lux-400lux is obtained with the optimization model of exposure time.Experiments take high-precision rack and cement pavement model under light intensity of 200lux as experimental measurement objects.Their results show that measurement with the optimized exposure time is more reliable.Mean residual sum of squares of the single light stripe measurement is about 0.03mm~2.The reconstructed model's average texture depth(MTD)of is only 9.8%away from the actual model.The results of experiments with cement pavement model under 80lux/200lux/300lux show that the measurement results with corresponding optimized exposure times are of good reliability and accuracy,and the values of MTD deviation(10.4%/9.8%/9.6%)are almost equal..(3)The quality of light stripe is influenced by different materials and different illumination conditions of drainage pipeline and pavement.The existing extraction algorithms cannot meet the high-precision light stripe center extraction(LSCE)for different quality images,which further affects measurement accuracy.Therefore,a novel light stripe center extraction algorithm based on Gauss-Lorenz peak fitting is proposed.The light stripes are firstly divided into four types according to image quality.The four traditional LSCE methods are introduced and compared with the extraction effects of the four types of light stripe.On the basis of the works above,the proposed algorithm is studied.Firstly,the energy model of light stripe is constructed and these models'composition of the four types of light stripe.According the characteristics of energy model,The Gaussian-Lorenz fitting model,including Gaussian main peak and Lorenz sub peak,is constructed.The two peaks can be used to fit principal component and sub component of the energy model.Finally,the accurate and reliable centers can be extracted with gray centroid method form the light stripe whose Lorenz sub component is removed from total energy distribution.In order to ensure the extraction accuracy and efficiency,related works,including image normalization,automatic extraction of LSROI and initial parameters setting of fitting model,are done before the extraction operation.The experimental results show that the extraction results of different quality light stripes with the proposed algorithm are the best.The average reliability value of the centers extracted with the proposed algorithm are bigger than that with others.It is time-consuming overall.However,when it is applied to images which are seriously influenced by diffuse and specular reflection,its time consuming is shorter than that of Steger algorithm which is considered as the best method to extract centers from the above light stripe images.(4)In order to verify the feasibility of research content,the application experiments of high-precision 3D detection with multi-vision line structured light for indoor drainage pipeline model and actual pavement are carried out.A system platform is set up,and a hardware system consisting of a measurement probe,a linear sliding table,a compensation information acquisition module and a communication module is developed.A software system composed of a probe control module,an image acquisition module and an image post-processing module is developed.An indoor experimental drainage pipeline model is constructed,whose inner-wall is fixed with five types of simulating defects with 3D significant characteristics.The 3D detection experiments of drainage pipeline model and cement and asphalt pavement are carried out.The application experiments obtained complete and accurate 3D information of these detection objects.The results verify the effectiveness of large-scale measurement method,exposure time optimization model and light stripe center extraction algorithm(5)Information of alien objects affect the accuracy of the drainage pipeline's information and further affect the reliability of follow-up evaluation results.Therefore,a removal method of alien object's information based on progressive morphological filter is proposed.Mathematical morphology is applied in this method and opening and closing operations with linear filter window are used to remove alien objects.In order to effectively remove variable alien objects,a new morphological filter with progressive window size is developed.Meanwhile,in order to avoid mistakenly removing their own information,a constraint condition,named elevation difference threshold,is added to the morphological filter.Finally,combining with the geometric characteristics of alien objects and their own features,parameters of the progressive morphological filter can be determined.The alien objects removal experiment of the 3D detection result of drainage pipeline inner-wall validates the effectiveness of the proposed method.Periodical results have been achieved in this theise.However,there are still many issues to be studied,such as rapid 3D detection,automatic identification of damages.