TPMT based Automatic Road Extraction from 3D Real Scenes

摘要

The Generation of digital 3D buildings and roads from a number of 3D real scenes is a very difficult and important task in modern city planning and management. A TPMT (Tilt Photographic Measurement Technique) based automatic road extraction method with 3D real sceneries is studied and demonstrated. Firstly, TPMT is used to obtain 3D data of the city and form its stereoscope image sequences. Secondly, Seamless stitching of the stereoscope image sequences and high-resolution remote sensing images are conducted to produce an integrated urban digital road feature database. Thirdly, image processing and pattern recognition methods are employed to identify and extract road from the urban digital road feature database and form a vector contour image of the road. Finally, the vector contours are classified with a pattern classifier. The theoretical analysis and practical demonstrations have shown that our approach has higher precision in automatic 3D road extraction from a number of 3D real scenes, which can meet the demand of unmanned driving for high-precision roads and 3D maps. Moreover, it significantly reduces the cost of road information acquisition compared to the existing method using Lidar and has potential to play a key role to promote and support ever growing urban planning, construction, management and emergency response. Faced with a big data problem in huge and various 3D real scenes, a deep learning approach is under consideration in our undergoing study and development.