Abstract: The deformation date of 3D geo-mechanics model are composed of the internal deformation of rock mass and convergence deformation of boundaries. Searching the method of measuring the convergence deformation with high precision has been major direction of research in the area of data acquisition for 3D geo-mechanics model. In this paper, the calculation method based on the coordinates transformation of images in the traditional digital camera photogrammetry was revised the small template operation method of 5 ? 5 was improved. Instead, the big templates with the layout of standard points was used and a method of selection for coordinate data of neighboring marked nods was designed, thus, the influence of lens distortion was reduced effectively. The direct linear transformation mechanism based on the BP neural network was used for coordinates recognition of images and comparison of coordinates of marked nods during experiment period. The measurement of convergence deformation for 3D geo-mechanics model was thus achieved. A 3D geo-mechanics model applying the close shot photographic technique of digital camera was set up for a large scale forked tunnels, the results of measurements were satisfactory.

Key words: convergence deformation, neural network, direct linear substitution, bifurcation tunnels, 3D geo-mechanics model test