Point cloud registration is necessary to acquire full-view data in coded SL three-dimensional measurement, based on single-view measured data. Aiming at surface feature of metal parts or human body, matching point pair construction principle, and transition matrix optimization method are analyzed. First, auxiliary-stereo-target registration principle and device are presented to establish matching point pair, and least-squares ill solution and iterative misconvergence caused by coplanar matching points can be avoided. Second, ICP method is adopted for acquiring transition matrix, and then mismatch point pair rejection method based on orthogonal Gray code principle is designed to increase iterative convergence. Experimental results show, registration error is about 0.8mm, close to that of global camera method and higher than that of surface method. This method has no influence on measured surface, and simplifies measurement device.