A three-dimensional topography measurement method for large-scale components is proposed to achieve high-precision and high-efficiency measurement. To address the issues of low accuracy and poor stability inherent in binocular structured light measurements， a complementary positive and negative Gray phase-shifted structured light coding technique is developed. This approach mitigates the susceptibility of binocular structured light systems to systematic errors and inaccurate phase unwrapping during actual measurement. The method employs a robust pixel classification technique for binarizing the positive and negative Gray codes， alongside an improved Gaussian filtering algorithm applied to the phase-shift code images. To improve the measurement accuracy of existing point cloud， an improved iterative nearest point cloud stitching algorithm is introduced. This algorithm extracts and filters point clouds from overlapping regions of adjacent perspectives to eliminate interference from non-overlapping areas and integrates the Levenberg-Marquardt optimization algorithm into the iterative process to enhance robustness against initial position sensitivity and noise interference. Compared to traditional iterative nearest point stitching methods， the proposed algorithm increases accuracy by 55%， accelerates stitching efficiency by several times， and reduces iteration count by 61%. Experimental results demonstrate that the developed system achieves length measurement accuracy better than 450 μm/m at a measurement distance of 700 mm and a camera angle of 65°， with point cloud stitching computation times of approximately 25 ms between adjacent views. These findings confirm that the method satisfies the accuracy and efficiency requirements for three-dimensional topography measurement of large-scale welded components.