Conventionally, for a direct- detection 3D imaging light detection and ranging (LIDAR) system, it′s impossible to directly and accurately determine the pixel-wise signal-acquisition time in the case that the scene properties are unknown. Thus, a time-of-flight depth sensor produces noisy range data due to different scene properties such as surface materials and reflectivity, frequently including either a saturated or severely noisy depth estimated. A photon-counting adaptive depth imaging strategy for 3D imaging LIDAR is presented. By improving the traditional imaging model based on the maximum likelihood estimation algorithm, an appropriate signalacquisition time is adaptively selected, and then determine an optimal depth for each pixel. Experiment demonstrates that the proposed algorithm can be more quick and more accurate to reconstruct the scene depth images even in the low light-level environment.