| The heterogeneous yields in the underground tunnel constitute a significant factor affecting normal underground operation. Therefore, it is of great urgence and necessity to build the instantaneous surveillance and safety assessment system on the long-term yields and deformantion in the tunnel, which enables us to apprehend the overall deformation, and to discover and eliminate potential safety hazard in time. Algorithm SLAM based on machine vision can restore linear characteristic of track space, and thus yield coordinate can be obtained through the comparison between the restored linear characteristic and the previous one. The core steps of SLAM consist of feature extraction, matching and three-dimensional reconstruction, which brings about the research contents of this paper——The Algorithmic Research and Implemnetation of Extract of Train Geometry Characteristic based on Machine Vision. The extracted geometric features of the track in the paper can provide technical support for the following Algorithm SLAM. The following contents are what have been researched in the paper:(1) From the perspective of matching, the geometric features of the picture will be extracted and transformed into feature study and track structure analysis extracted by linear features. In the course of picture processing, we will study Gauss Noise Reduction through Filtering, gray-scale transformation of the picture, binarization of adaptive threshold value, tunnel shape filling and constructed function in order to conduct morphological closed operation, negation and researches of other theories.We will also analyze several kinds of edge detection formula, and obtain Canny Detection Formula proper to the paper proceeding from theories and experiments.(2) We have analyzed a standard Hough Transformation of the line extraction algorithm. Since Hough Transformation may occupy much of RAM, we have intensified the research on Hough Transformation. Furthermore, we have improved the random Hough Transformation based on its demerit of extracting pseudo pixel. Moreover, we have analyzed matching approaches of track lines based on linear geometricalproperties. With regard to the blind matching, an improved algorithm has been developed which takes advantage of the rough matching prior to the precise matching.The rough matching is based on linear portion length, gradient direction and centerpoint position, while the precise one focuses on the gradient information of the center point of the linear portion. Finally, the accuracy of the algorithm will be verified.(3) The establishment of hardware developing platform with Laser Sensor ACR-LD131, FPGA, and DSP Demoboard as its core, including the design of DSP Circuit Board,substantial data processing of laser sensor by DSP, which triggers the camera to take photos and the image pre-processing of FPGA.At the end of the paper, the defects of the feature extraction, match, hardwaredesign work and the deficiency of image preprocessing by FPGA in the paper havebeen summarized.Key points of the paper:(1) Linear feature extraction of the picture—RHT;(2) Linear feature extraction of the picture based on line segment attributes and midpoint descriptor of;(3) The image pre-processing of FPGA and DSP data processing of laser displacement sensor. |
