Abstract: Single-photon lidar has been widely used to obtain depth and intensity information of a three-dimensional scene. For multi-surface targets, such as when the laser transmit through a translucent surface, the echo signal detected on one pixel may contain multiple peaks. Traditional methods cannot accurately estimate multi-depth images under low photon or relatively high background noise levels. Therefore, a time-correlated multi-depth estimation method was introduced. Based on the time correlation of the signal responses, a multi-depth fast denoising method was adopted to point cloud data, and could identify the signal responses of multiple surfaces from background noise on each pixel. Considering the Poisson distribution model of the signal response set, the spatial correlation between pixels was introduced through total variation (TV) regularization to establish a multi-depth estimation cost function. The fast-converging alternating direction method of multipliers (ADMM) was used to estimate the depth image from the cost function. Experimental results on a multi-depth target at a distance of about 1 km show that the root mean square error (RMSE) and signal to reconstruction-error ratio (SRE) of the depth image estimated by the proposed method can be at least 27.05% and 18.39% better than that of other state-of-the-art methods. In addition, the data volume of this method is reduced to 4% of the original. It is proved that this method can effectively improve the multi-depth image estimation of single-photon lidar with smaller memory requirements and computational complexity.