Periodic Reporting for period 1 - MorE3D (Morphological Entities detection and characterisation from 3-D laser scanned point-clouds)
Período documentado: 2021-03-01 hasta 2023-02-28
Landscape reshaping processes have an immense effect on humans, being a fundamental component of their habitat. The morphological signatures they leave on the terrain enable us to trace them, understand their nature, develop strategies to avert hazards and provide a more sustainable future. As such entities are better traced in their natural 3-D shape, the last decade has seen an expedite growth in the use of high-resolution laser scanning (LiDAR) technologies to document, monitor, and analyse them. Nonetheless, the unorganised nature, span and massive data volume, turn the interaction with the acquired data cumbersome and difficult. Hence, common practices are rooted in manual feature delineation or in use of off-the-shelf raster-based tools, which were developed for different applications and scales. The outcome may be subjective and prone to misidentifications or distortions. As there is an evident gap between the richness of the data and geoscientists practices, I propose in MorE3D to develop a new processing framework that strengthens that link. This will enable to highlight features, provide quantitative morphometric information, and facilitate analysis of trends and patterns which are essential to asses natural processes. Based on the understanding of the geometric signature recorded within the data, I propose to cast this as an energy-based approach that uses global optimisation to detect entities. Furthermore, as many applications combine active and passive sensors (lasers and cameras, for example) or use drone-based imaging to supplement the data during acquisition, I will to further extend the proposed scheme and develop a unified multichannel optimisation framework, where all acquired information is integrated and utilised. Such models will open new avenues to analyse features, detect patterns, trace changes, and essentially enable accurate measurements of the processes affecting landforms, while paving the way to the relevant research communities.
In this action, we have developed automated and reliable techniques for accurately characterizing geomorphic phenomena. By analysing the geometric patterns present in the data, we employed an energy-based approach that utilizes global optimization to identify these entities. We detected geometric cues for the extraction through a deep-learning scheme that learns the topography and recognizes anomalies therein. External cues, such as colour or signal intensity, are also integrated into a cohesive model. Lastly, to allow a focused search of patterns, geometric constraints are added to the general model.
The proposed methodology was developed in three main stages. We first developed the 3D extraction framework of geomorphic features, including feature based and learned geometric cues. Then, external information was integrated to the model. Eventually, geometric constraints were introduced. Each stage was tested on laser scans acquired by various platforms and sensors. To ensure the generality of the developed method, a large variety of datasets and scenes were used, where different types of geomrphic phenomena existed. Dissemination to the geoscientific community was carried out through relevant workshops and conferences, where the developed tools were presented. In these events, we focused on creating new collaborative opportunities. As an outcome, we applied the proposed framework on dynamic processes. Reaching out to larger audiences, MorE3D was presented in events that aim to expose the public to research work (e.g. European Researchers Night). To that end, advertising materials were created, distributed, and presented.
The proposed methodology was developed in three main stages. We first developed the 3D extraction framework of geomorphic features, including feature based and learned geometric cues. Then, external information was integrated to the model. Eventually, geometric constraints were introduced. Each stage was tested on laser scans acquired by various platforms and sensors. To ensure the generality of the developed method, a large variety of datasets and scenes were used, where different types of geomrphic phenomena existed. Dissemination to the geoscientific community was carried out through relevant workshops and conferences, where the developed tools were presented. In these events, we focused on creating new collaborative opportunities. As an outcome, we applied the proposed framework on dynamic processes. Reaching out to larger audiences, MorE3D was presented in events that aim to expose the public to research work (e.g. European Researchers Night). To that end, advertising materials were created, distributed, and presented.
For the extraction of geomorphological entities, region-based level sets approach was brought together with gridded Lidar data. These were further developed in MorE3D with PHOTO expertise in point clouds to formulate and implement the level sets approach on point cloud data for the detection of morphological features while using learned saliency as internal cues. In this way, we have provided a holistic extraction framework for landscape entities and processes which does not require prior information, is not limited to the type or number of entities in the scene, and are unrestricted to a specific point in time.
In this regard, MorE3D provided new ways to quantitatively describe morphological phenomenon and characterise its shape. The integration of multiple channels is unique and facilitates an efficient way to extract entities according to attributes that are not necessarily geometric. Therefore, MorE3D enables advanced tools to understand the landscape processes, their past and current phase, and thus it facilitates better management of Earth’s natural systems, all while attending to environmental values and promoting a sustainable development.
In this regard, MorE3D provided new ways to quantitatively describe morphological phenomenon and characterise its shape. The integration of multiple channels is unique and facilitates an efficient way to extract entities according to attributes that are not necessarily geometric. Therefore, MorE3D enables advanced tools to understand the landscape processes, their past and current phase, and thus it facilitates better management of Earth’s natural systems, all while attending to environmental values and promoting a sustainable development.