In the circumstances of low luminance, visible assistant light is not available or convergence is very similar to the background or syncretism to each other, research on the local 3-D image restoration method is of great sense. Generally to restore 3-D image from binocular 2-D image, active vision is in need. Otherwise the application would be only constrained in objects of simple shape. A local 3-D image restoration method based on infrared grid raster was proposed in the paper. Its principal is irradiating the goal object by non-structure infrared grid raster, catching the object area image by double CCD infrared camera and distilling the grid data from the image. Then the 3-D grid image of the target was restored by the binocular image matching from the grid data. Finally, the practice 3-D image of the object is fitted from the 3-D grid image. By using the infrared grid raster, the target could be added with additional grid without frightening it and the process was quite efficient. At present, the stereo image restoration based on grid and grid raster was realized. Thus, through optimizing the existing image pretreatment technology, character distilling technology and stereo matching technology, the clear local 3-D image of the target would be restored faster even in the case of lower environment luminance while the visible assistant light is forbidden.