Length of the wall is about 150 meters (~490 ft). Made from 125 photos using a 12MP camera 24mm kit lens while walking and not paying much attention to the focus and angle. Really shows how powerful SfM algos are nowadays and I haven’t even used the sequential workflow. The resizing factor and the Hessian threshold used here are smaller than the ones used in the previous models, resulting in a greater number of features in each image. Feature-based matching is applied only to the overlapping images. A cross-check test is implemented, according to which the similarity of the descriptors is verified through reverse comparison using the criterion of the minimum Euclidean distance, which is constrained to be below a predefined threshold. The RANSAC algorithm is applied for the removal of outliers through the estimation of the fundamental matrix. Optimized from 2.7M vertices.
Comments