Photogrammetric multisensor application for the documentation of the Underwater Heritage

Department: PhD in Architectural and Landscape Heritage
Supervisor: Prof. Filiberto Chiabrando
Candidate: Alessio Calantropio

In the last decade, deep learning (DL) revolutionized artificial intelligence in almost all its fields of application. Indeed, it allowed to reach impressive results in both perception (computer vision, natural language processing) and decision making (reinforcement learning, navigation). This drastic change was aided by a huge boost in computing capabilities, and by the development of components specifically dedicated to massive data processing.

However, robotic systems usually lack the computational power typically dedicated to deep learning algorithms due to cost and dimensions. Nevertheless, control applications (like autonomous navigation, for example) must give particular attention to latency and power consumption, as the slightest delay could cause severe damages. Due to these challenges, model optimization and edge AI have become of enormous interest in the last few years, as embedding such powerful computing closer to sensors and actuators is the key for the future of intelligent robotics.

For this reason, my research focuses on the development of deep learning models to obtain intelligent sensors for robot perception, control and decision making. Particular attention is given to the constraints imposed by the application field in terms of latency, power consumption and weight.

Although underwater photogrammetry has become widely adopted, there are still significant unresolved issues – especially those related to the acquisition and the processing phases of underwater imaging – that are worthy of attention. Due to the need for generating an accurate virtual 3D replica or twin of the surveyed object or site, it is necessary to address certain issues, such as preserving consistent radiometry, avoiding blurry and low-contrast or over/under-exposed images, and the like. It is possible to identify three main topics regarding the issues connected to photogrammetric applications in underwater environments: a first point concerns the generation of the 3D models and the related metric products, which is related to best practices and standards adopted during the acquisition phase (e.g. coverage, camera calibration, radiometric correction, integration with data from different sources, etc.). A second point is related to the correct georeferentiation of the generated 3D model and the related metric products, due to the impossibility of relying on GNSS navigation systems. Last, but not least, a third topic is related to dissemination and management, as certain visualization platforms offer the possibility of interacting with the model (with features like generating geometric sections, writing dynamic annotation, etc), and sharing models in ways that facilitate communication within or between different user groups. To solve the problem of the impossibility of relying on GNSS systems and to acquire the coordinates of control points, in acquisitions with an optical sensor cases, it is possible to use an integrated underwater positioning system GNSS and Short Baseline Acoustic positioning system to obtain set-up and position of the sensor related to the product to be detected.