Volume Estimation for a Building Based on Aerial Images and Point Cloud Modeling

On some occasions and for some purposes, we need to estimate the volume of a large object, such as a building, a statue, an airplane, etc. This paper uses the aerial image captured by a low-cost standard drone and the point cloud model established by an open-source tool to achieve a large object volume estimation. The proposed method combines high-resolution point cloud models generated via RealityCapture and some data preprocessing to start the task. The data preprocessing includes DBSCAN clustering for noise removal and height-based filtering to refine the object’s outlines. In addition, this method improves accuracy by using an algorithm that adjusts the dilation kernel to compensate for sparse areas in the point cloud. “Adjusting the dilation kernel” primarily refers to dynamically altering the size of pixel dilation within the image to restore fractured contours caused by sparse or scattered point cloud projections. This process helps fill in edge details in each slice image and reduces estimation bias due to insufficient points, thereby enhancing the overall accuracy of the volume calculation. We chose the high building in Taipei City as the experimental object to verify the effectiveness and accuracy of the proposed method.