Point Clouds Denoising Method Based on SPAD Discrete Probability Model and Completion

The laser imaging system based on single-photon avalanche diodes (SPADs) has demonstrated significant promise in long-range or low-power detection. However, the obtained point clouds encompass intricate noise attributable to the device’s detection characteristics, significantly affecting downstream applications. In this article, we present a novel denoising method named LCLOF for SPAD-acquired point clouds, built upon a completion denoising strategy and the SPAD discrete detection probability model. Instead of directly denoising the input raw data, LCLOF outputs clean data through point cloud completion. Specifically, the proposed method involves preprocessing the input data to extract feature points and incomplete point clouds. The SPAD discrete probability model is integrated into point cloud preprocessing to achieve the desired outcome. Moreover, an improved local outlier factor (ILOF) approach is put forth as the completion strategy. Denoising will be accomplished by using ILOF to filter out feature points and fill the incomplete point clouds. Theoretical analysis illustrates the feasibility of the completion strategy in scenarios involving point clouds with nonuniform density. Experimental results illustrate the performance of the proposed method over state-of-the-art approaches concerning the denoising of point clouds in the context of SPAD-based laser imaging system.