Automatic feature type selection in digital photogrammetry of piping

A building information model for pipes already in place is essential in maintenance, for example, mending, reconstruction, and modernizing. However, current point cloud construction methods are not suited to complex piping systems, and many point cloud merging methods perform poorly in complex piping environments. To provide critical functions for constructing pipe point clouds from digital photogrammetry, a feature type selection network (FTSNet) is proposed. First, a digital photogrammetry method combining FTSNet with handcrafted features is proposed. Second, a piping data set of various piping scenes is constructed and evaluated using the developed data capturing device. A training data set and the corresponding network output categories are determined according to pose‐estimation performance using different image feature types. Finally, experiments conducted on the final test data set indicate that, together, digital photogrammetry and FTSNet can improve accuracy, flexibility, and processing speed.