Abstract: Dense mapping estimation is an important goal for simultaneous localization and mapping (SLAM). Considering the poor reconstruction accuracy of the depth filtering algorithm, an improved monocular dense point cloud map reconstruction method based on inverse-depth filtering is proposed. This algorithm improved the efficiency in the epipolar search phase by setting threshold, and used the inverse depth Gaussian filter to update posterior inverse depth probability distribution. The outside points were eliminated through intra-frame detection. Experimental results show that the improved dense reconstruction method has denser, more accurate reconstruction effects without GPU acceleration.