Inland navigation is a prime transportation vector, whether for freight or tourism, and represents a more environmentally friendly alternative to road transport. The maintenance of fluvial infrastructures is not only a matter of heritage preservation, but also a commercial necessity and a security issue. Their documentation and periodic inspection is challenging due to the variety and huge number of structures to be considered. For example, Voies Navigables de France (VNF, the French operator of waterways), manages more than 6700 km of waterways (including more than 4000 structures). Hence, filling in the VNF structure database (called BDO) which, for the moment, is performed through systematic and exhaustive technical surveys of structures, carried out on site by agents, is a major effort. For this reason, it is necessary to develop highly efficient methods that are minimally invasive to fluvial traffic and require as little human intervention as possible.
In this presentation, we focus on a particular type of structures, namely canal tunnels. Although there are only a small number of them (e.g. in France, 33 tunnels are still in operation, for a total of 42 km of underground waterways), they are key to safe navigation. Located mostly on small gauge canals, they attract heavy touristic traffic. We illustrate how up-to-date technologies (3D reconstruction by photogrammetry, high frequency microbathymetry, pattern recognition techniques) may be used to design automatic tools for the detection of damages on structures, and to produce accurate 3D models including both under- and above-water parts of the tunnels, even in the absence of GPS signal.
In 2009, a French partnership composed of VNF, The Centre d’études des Tunnels (CETU) and the CETE de l’Est (now the Cerema) in collaboration with the Photogrammetry and Geomatics Group (INSA) has developed a visual inspection system dedicated to canal tunnels from acquisitions of image sequences. A modular prototype, mounted on a barge, has been devised for recording images of the vaults and sidewalls of the canal tunnel. Pattern recognition algorithms have been implemented for automatically detecting defects in the infrastructure. We rely on supervised learning classification algorithms from different representations of the data used at the classifier entrance. As first results, we have shown that cracks, water leaks, exposed steel reinforcements and damaged masonry joints can be detected by the algorithm, which is also able to disctinguish several kinds of linings (e.g. masonry, concrete, rock).
With regard to 3D modeling, Emmanuel Moisan's doctoral work (funded by a Cerema scholarship from 2014 to 2017) allowed us to explore the whole-tube 3D imaging of canal tunnels by combining photogrammetry and high-frequency microbathymetry sonar surveying. More specifically, three aspects were studied:
- the evaluation of a recent technology in 3D sonar imaging: We took advantage of the emptying of a lock for maintenance purposes to survey it with a Terrestrial Laser Scanner (TLS). Once the lock was back in operation, a sonar inspection was carried out. For this purpose, a Multi-Beam EchoSounder (MBES), namely Blueview BV5000 was operated in a similar way as the TLS, using a Mechanical Scanning System (MSS). While 3D reconstruction from TLS data is straightforward, a specific processing pipeline had to be devised for dealing with sonar measurements. Finally the models issuing from TLS and MBES data were compared. The study highlighted the artifacts that must be dealt with in shallow and narrow environments. It also shown that defects as small as 5 cm may be detected using this technology.
- the development of a 3D reconstruction chain based on data acquired by a boat: Since GPS signals are not available in tunnels, traditional bathymetric techniques cannot be used to register the 3D profiles provides by the MBES along the trajectory of the boat. To overcome this difficulty, we proposed to take advantage of the properties of photogrammetry, which is a technique that reconstructs a 3D model automatically from a series of images of the work, and also provides the localization of these images. It is then possible, having previously calibrated the relative position of the cameras and sonar on the boat, to estimate the trajectory of the MBES. Finally, the 3D model resulting from photogrammetry and the series of 3D sonar profiles positioned perpendicular to the trajectory are juxtaposed to form the complete 3D model of the canal tunnel.
- the quantitative evaluation of the 3D model against a reference model of the structure built from static measurements: A survey of the entry of a canal tunnel was performed by combining TLS and MSS technologies. Here again, an original method had been proposed to localize the sonar scans taken from unknown static positions. The comparison shows that the resolution of the above-water model is millimetric (with centimetric accuracy), and that the underwater model resolution is centimetric, with decimetric accuracy.
Even if steps still need to be taken before a complete automation of the process can be achieved, this research has shown the feasibility of documenting engineering structures such as canal tunnels in 3D. The proposed techniques could also be generalized to areas where the GPS signal is inoperative. They are complementary to more classical bathymetry methods. Moreover, we have shown the feasibility of automatically detecting components or defects in the structure via pattern recognition methods, which may be generalized to other structures and to underwater elements.
In line with this collaborative research, and as part of its evolution towards digital technology, the so-called RES' O FLUVIAL 2.0 project, VNF is launching a first innovation partnership for high-throughput collection of images of the linear components of its network (mainly embankments and dikes), from the waterway. The aim is to automatically pre-inform (using shape recognition technologies) the condition field in the VNF structure database. Moreover, the images will be made available through a geolocalized interface (similar to Google Riverview).