For the quality of robot polishing process, the laser scanning technology is applied to the measurement and assessment of the shaping error and the clamping error of robot clamping workpiece, including the acquisition and denoising process of point cloud data. A stripe type laser scanner in uniform linear motion is used to scan a part clamped by robot. The approximate grid point cloud is obtained by adjusting the measurement and motion parameters. In order to remove the large scale noise in point cloud, the local mean K-nearest-neighbor mean filter (LMKMF) based on the K nearest neighbor mean filter(KNNMF) is proposed as local filter of partial large scale data point in advance. Relevant mathematical model is established. Peak signal-to-noise ratio is used as evaluation standard, and the actual measurement point cloud samples are used as the denoising test object. Results show that compared with the KNNMF, the denoising ability of the algorithm combining LMKMF with KNNMF has an improvement of 53.78% under the noise density of 30%, which proves that the proposed algorithm has greater ability of denoising and detail preserving when the noise density is high.