3D point cloud maps are critical for self-driving cars and robots to perform navigation and localization in complex outdoor environments. However, lidar scans used for map construction may contain dynamic objects which leave ghost trail in the map. The trail will inevitably deteriorate the quality of maps and affect both localization and navigation accuracy. Therefore, it is necessary to remove dynamic objects from the map. However, state-of-the-art approaches are susceptible to noise and changes in incident angle, which can result in the unintended removal of non-dynamic points. Additionally, due to occlusions, dynamic objects may not be completely removed from the point cloud. To tackle the problem, this paper presents a novel static map constructing method called DynamicRemover. Unlike previous works that eliminate dynamic objects at the point level, our method focuses on calculating the probability of an object being dynamic. Initially, we employ a visibility-based algorithm to obtain a suspected dynamic point cloud, comprising points or regions that may belong to dynamic objects. Subsequently, we analyze the distribution of each object in both the suspected dynamic point cloud and the original point cloud to calculate its probability of being a dynamic object. Experimental evaluations carried out on the SemanticKITTI dataset demonstrate that our method exhibits more robustness and reliability compared to other approaches.