煤炭工业是我国最重要的基础能源产业，矿难事故的频发引起了国家和社会对煤矿安全问题的高度重视。随着视频监控技术的快速发展，视频监控技术已在煤炭安全生产监测中得到广泛应用。本文主要研究了立体视觉测量中图像特征点提取、摄像机标定和重建点坐标求解三个关键部分，并实验应用立体视觉测量方法监测煤矿巷道形变。主要研究内容和成果如下： 首先研究了图像上角点和光斑中心像点的提取方法。实验对比了Harris、Shi-Tomasi和FAST三种角点检测算法，结果表明：Harris算法对阈值敏感且易产生角点聚簇；FAST算法易提取到伪角点和产生角点聚簇；Shi-Tomasi算法能使检测到角点分布均匀，避免角点聚簇，得出Shi-Tomasi算法适合十字靶标角点像素级坐标提取。通过图像预处理由Canny边缘检测算法得到光斑单像素边缘轮廓，利用最小二乘拟合法获取光斑中心像点坐标。 其次对摄像机平面模板标定法进行了研究。由于通常计算单应性矩阵时使用了畸变较大的像点影响了内部和外部参数初值，从而影响了标定精度。本文提出在计算单应性矩阵时仅使用图像中心附近像点，使内部参数和外部参数初值能更好地逼近准确值，提高摄像机标定精度。实验证明此方法能提高摄像机标定精度。 然后针对立体视觉空间点坐标求解问题，分析比对了三种常用求解方法：最小二乘法、公垂线段中点法、基于极线约束的重建方法，给出了以公垂线段为约束使反投影像点误差最小的空间点坐标求解方法，实验证明该方法能提高空间点坐标求解精度。 最后将立体视觉测量技术应用到巷道变形监测中，给出了一种巷道变形监测实施方案，实验对巷道上监测点位移进行了测量，结果证明该视觉测量方案能准确计算出巷道上监测点空间位移，能对巷道变形进行监测。

Coal industry is our country most important foundation energy industry, and the frequent occurrence of coal mine accidents makes the coal mine safety has become highly concern of the state and society. With the rapid development of video surveillance technology, video surveillance technology has been widely used in coal safety production monitoring. This paper has mainly studied three key components, which includes the extraction of the image feature in stereo vision measurement, camera calibration and solving rebuilding coordinates, and taken advantage of stereo vision measurement to monitor the deformation of coal mine tunnel. The main research content and results are as follows: First of all, the paper has studied the extraction method of corner and spot center dot in image. Experiment compared the three kinds of corner detection algorithm, including the Harris, Shi-Tomasi and FAST. The results show that the Harris algorithm is sensitive to the threshold and easy to produce angular point clustering; FAST algorithm is easy to detect false corner and produce angular point clustering；Shi-Tomasi algorithm can make the Angle of the detected points distributed evenly, avoiding Angle of point cluster, it is concluded that Shi - Tomasi corner detection algorithm for the target angular point coordinates extraction pixel level. Spot is obtained by image pretreatment by Canny edge detection algorithm of single pixel edge profile, then extracted the spot center image point by the least squares fitting method. Secondly, the paper has studied the camera calibration plane template method, because using the greater variable pixel in computing homography matrix has affected the initial value of internal parameters and external parameters, calibration precision has been affected. In order to make the initial value of internal parameters and external parameters preferably come near accurate value, the paper only put to use the pixel nearby picture centring in computing homography matrix, thus a better camera calibration precision has been got. Experiments prove that this method can improve precision of camera calibration. Then aiming at solving the point coordinates in stereo vision space, the paper has analyzed three kinds of commonly methods: least squares method、the male in the vertical section point method and based on the reconstruction of the polar constraint method. This paper has put forward to male vertical section for a constraint to the projection image point space point coordinates, solving method of minimum error, experiments has proved that this method can improve precision of camera calibration. Finally, the stereo vision measurement technology is applied to the roadway deformation monitoring, and a roadway deformation monitoring implementation scheme is given, monitoring stations displacement on the roadway were measured, The results have showed that the visual measurement scheme calculate accurately the displacement of monitoring space and monitor on roadway deformation.