According to the data of space target surface reconstruction of TOF depth camera point cloud based on source vulnerable to the instrument itself, scanning the environment, the effect of external disturbances, invalid points and noise points and large amount, increases the computational burden and the impact of the reconstruction quality problems, this paper proposes a denoising method of random sample consensus background segmentation of point cloud based on knearest neighbor, to eliminate outliers and invalid target data. Firstly, the improved RANSAC (random sample consensus,RANSAC) algorithm, by setting different threshold of the original point cloud background segmentation, to Ensure the accuracy of extracting the main features of the reconstructed object. Then, through the Knearest neighbor point cloud average algorithm and bilateral filtering algorithm to remove outliers, finally using voxel grid method to achieve point cloud data sampling, simplified target point cloud, retains the local characteristics, to speed up the reconstruction speed. The experimental results show that the the algorithm can effectively eliminate noise, high accuracy, good realtime performance, meet the application requirements.