Photogrammetric Techniques and Image Segmentation via Machine Learning as Supporting Tools in Paving Asphalt Mixtures Studies

Research activities in the field of bituminous aggregates composition for roadway paving (asphalt) often require 2- and 3-D geometric measurements. In some cases, these can provide support in studying aggregate arrangement within bituminous matrices subjected to stress and evaluating 3D trends in the tyre/road contact interface. The current job investigates the potentialities of photogrammetry in handling these issues, presenting and checking two methodologies, i.e. the tested and well-established homographic transformation for 2D investigations, and the more innovative, 3D approach of Structure from Motion (SfM) + Multi-View Stereo (MVS). Both theoretical provisions and test findings confirm that photogrammetry can offer effective solutions to surveying issues in this specific field. Accuracy of geometric measures and photographic quality are appropriate for investigating aggregate dimension and dislocations ranging at 10−4 m. Finally, some possible applications of Machine Learning (ML) algorithms for segmentation and classification of photogrammetry imagery are discussed. Automatic image segmentation leads to less than 2% mismatch in the interest classes.