Optical non-contact measurement of large-scale scenes has wide application in engineering. The multi-view reconstruction based on a single camera needs to arrange scale datum into the measurement area, and the image acquisition process is complex and requires a high overlap ratio between adjacent images. Therefore, a method of multi-view reconstruction using binocular cameras was presented. This method not only inherits the advantages of large measurement range in multi-view reconstruction, but also does not need to arrange scale datum, and the image acquisition process is simpler. Besides，the double camera constraint was added in the process of data optimization. Specific methods are as follows: firstly the internal parameters and solid connection of two cameras installed on the fixed rod were obtained according to camera calibration; then the binocular cameras were used to acquire the images of the area to be measured and obtain the spatial point cloud by binocular intersection. The transformation relation of the coordinate system of point cloud in adjacent time was calculated by the method of pose estimation, and the camera′s parameters and point cloud′s coordinate of each moment were transformed to the specified coordinate system. The position and pose of the camera and the coordinate of point cloud were optimized by using the bundle adjustment, and the size parameters and the deformation information were calculated from the point cloud coordinates. The test results show that in the range of 5 m, the measurement′s error is 3 mm, the average error is 1 mm. This method is suitable for the measurement of large scale object and scene and is stable and reliable.