Abstract:In the reconstruction process of the single photon LiDAR 3D depth image, the linear/exponential fitting method takes a long time to process and the signal-to-noise ratio of the reconstructed 3D image is not high, which increases the difficulty of subsequent target detection. Therefore, an improved fitting method based on the full-image statistics is proposed in this paper. In the spatio-temporal domain, global noise is used to improve the fitting effect of the noise function, which can enhance the signal-to-noise ratio and real-time performance of the depth image. Outdoor experimental results show that, with only 60 frames, the proposed method improves the target recovery rate by 41.21% and increases the signal to noise ratio (SNR) by 3.23 times compared to traditional fitting methods, while also significantly optimizing computational efficiency. Based on the reconstruction of the depth image, median filtering is used in the spatial domain to improve the signal-to-noise ratio of the depth image. Furthermore, the 3D connected component labeling algorithm is adopted to expand the target extraction method based on the connected component labeling from a binary image to a 3D depth image, which can improve the representation ability of the target features in the depth image as well as the distinguishing capacity for targets of different depths. The results show that the improved fitting method based on the global noise and the target extraction method based on the 3D connected component labeling have good signal-to-noise ratio improvement capabilities and 3D depth image target feature extraction capabilities.