Abstract There are two problems that are regard as a bottleneck in the traditional geometry reconstruction by using a coordinate measuring machine. This is because that the geometric information lacks of before the digitizing process. (1) A fixed measuring span is adopted for the measuring strategy. (2) It cannot offer radius compensation for point data correctly and immediately. A photometric stereo technology based on neural network is presented for automatic acquisition process with adaptive measuring spans as well as correct and immediate probe radius compensation according to the variations of curvature on the measured surface. Experimental results demonstrate that the integration of the acquisition process with the photometric stereo technology makes it possible to complete geometry reconstruction at a time.