Panagiotis Partsinevelos

Semantic segmentation plays a significant role in unstructured and planetary scene understanding, offering to a robotic system or a planetary rover valuable knowledge about its surroundings. Several studies investigate rover-based scene recognition planetary-like environments but there is a lack of a semantic segmentation architecture, focused on computing systems with low resources and tested on the lunar surface. In this study, a lightweight encoder-decoder neural network (NN) architecture is proposed for rover-based ground segmentation on the lunar surface. The proposed architecture is composed by a modified MobilenetV2 as encoder and a lightweight U-net decoder while the training and evaluation process were conducted using a publicly available synthetic dataset with lunar landscape images. The proposed model provides robust segmentation results, allowing the lunar scene understanding focused on rocks and boulders. It achieves similar accuracy, compared with original U-net and U-net-based architectures which are 110–140 times larger than the proposed architecture. This study, aims to contribute in lunar landscape segmentation utilizing deep learning techniques, while it proves a great potential in autonomous lunar navigation ensuring a safer and smoother navigation on the moon. To the best of our knowledge, this is the first study which propose a lightweight semantic segmentation architecture for the lunar surface, aiming to reinforce the autonomous rover navigation.

Feature extraction plays a crucial role in computer vision and autonomous navigation, offering valuable information for real-time localization and scene understanding. However, although multiple studies investigate keypoint detection and description algorithms in urban and indoor environments, far fewer studies concentrate in unstructured environments. In this study, a multi-task deep learning architecture is developed for keypoint detection and description, focused on poor-featured unstructured and planetary scenes with low or changing illumination. The proposed architecture was trained and evaluated using a training and benchmark dataset with earthy and planetary scenes. Moreover, the trained model was integrated in a visual SLAM (Simultaneous Localization and Maping) system as a feature extraction module, and tested in two feature-poor unstructured areas. Regarding the results, the proposed architecture provides a mAP (mean Average Precision) in a level of 0.95 in terms of keypoint description, outperforming well-known handcrafted algorithms while the proposed SLAM achieved two times lower RMSE error in a poor-featured area with low illumination, compared with ORB-SLAM2. To the best of the authors’ knowledge, this is the first study that investigates the potential of keypoint detection and description through deep learning in unstructured and planetary environments.

Phytoliths constitute microscopic SiO₂-rich biominerals formed in the cellular system of many living plants and are often preserved in soils, sediments and artefacts. Their analysis contributes significantly to the identification and study of botanical remains in (paleo)ecological and archaeological contexts. Traditional identification and classification of phytoliths rely on human experience, and as such, an emerging challenge is to automatically classify them to enhance data homogeneity among researchers worldwide and facilitate reliable comparisons. In the present study, a deep artificial neural network (NN) is implemented under the objective to detect and classify phytoliths, extracted from modern wheat (Triticum spp.). The proposed methodology is able to recognise four phytolith morphotypes: (a) Stoma, (b) Rondel, (c) Papillate, and (d) Elongate dendritic. For the learning process, a dataset of phytolith photomicrographs was created and allocated to training, validation and testing data groups. Due to the limited size and low diversity of the dataset, an end-to-end encoder-decoder NN architecture is proposed, based on a pre-trained MobileNetV2, utilised for the encoder part and U-net, used for the segmentation stage. After the parameterisation, training and fine-tuning of the proposed architecture, it is capable to classify and localise the four classes of phytoliths in unknown images with high unbiased accuracy, exceeding 90%. The proposed methodology and corresponding dataset are quite promising for building up the capacity of phytolith classification within unfamiliar (geo)archaeological datasets, demonstrating remarkable potential towards automatic phytolith analysis.

Precise coordinate estimation is a fundamental engineering challenge in many mapping applications. Conventional surveying techniques based on global navigation satellite system (GNSS), light detection and ranging (LiDAR), or total stations involve costly equipment and time-consuming methodologies and may suffer from restrictions, such as occlusions and low satellite availability. In this study, a surveying approach, based on a custom-equipped unmanned aerial vehicle (UAV) and ArUco markers distributed in an unknown area as the potential target mapping points, is proposed and evaluated through simulation. The UAV incorporates a real-time kinematic GNSS receiver and a gimbal unit, with a simple camera and an electronic rangefinder module. The system demonstrates a real-time hierarchy targeting system that allows the UAV to engage with the ground targets through the camera, measure corresponding distances, record UAV coordinates, and then perform a multilateration-based target coordinate estimation. To evaluate the flexibility, efficiency, and onboard performance of the proposed target positioning approach, the method was developed as a robotic operating system software package and tested in the Gazebo robotic simulator on an NVIDIA Jetson TX2. Several mapping environments, along with varying flight type scenarios, were created to evaluate the resulting coordinate estimation errors. The achieved positioning accuracy is very promising and demonstrates a possible use for circular flying trajectories. Along these lines, the proposed methodology may pave the way toward a precise surveying alternative, as it may establish the main surveying method for common mapping applications in the near future and even provide coordinate estimations in demanding and, until now, unattainable areas.

Recent progress on geospatial and sensory, artificial intelligence technologies defines the necessity to revisit conventional geodetic techniques for surveying and mapping. In the present study, an alternative novel surveying method is implemented, which enables the precise localization of characteristic points in any area, including unknown and GNSS-denied environments, by simply using low-cost cameras. The methodology is based on novel algorithms that combine simultaneous localization and mapping (SLAM), deep learning, point-cloud processing, along with coordinate systems’ transformations. The camera system subsequently detects and localizes target markers and reconstructs a 3D environment with relative coordinate estimations under a few centimeters-level of accuracy.

Uncrewed aerial vehicles (UAVs) are continuously gaining popularity in a wide spectrum of applications, while their positioning and navigation most often relies on Global Navigation Satellite Systems (GNSS). However, numerous conditions and practices require UAV operation in GNSS-denied environments, including confined spaces, urban canyons, vegetated areas and indoor places. For the purposes of this study, an integrated UAV navigation system was designed and implemented which utilizes GNSS, visual, depth and inertial data to provide real-time localization. The implementation is built as a package for the Robotic Operation System (ROS) environment to allow ease of integration in various systems. The system can be autonomously adjusted to the flight environment, providing spatial awareness to the aircraft. This system expands the functionality of UAVs, as it enables navigation even in GNSS-denied environments. This integrated positional system provides the means to support fully autonomous navigation under mixed environments, or malfunctioning conditions. Experiments show the capability of the system to provide adequate results in open, confined and mixed spaces.

Renewable Energy Sources provide a viable solution to the problem of ever-increasing climate change. For this reason, several countries focus on electricity production using alternative sources. In this paper, the optimal positioning of the installation of wave energy converters is examined taking into account geospatial and technical limitations. Geospatial constraints depend on Land Use classes and seagrass of the coastal areas, while technical limitations include meteorological conditions and the morphology of the seabed. Suitable installation areas are selected after the exclusion of points that do not meet the aforementioned restrictions. We implemented a Deep Neural Network that operates based on heterogeneous data fusion, in this case satellite images and time series of meteorological data. This fact implies the definition of a two-branches architecture. The branch that is trained with image data provides for the localization of dynamic geospatial classes in the potential installation area, whereas the second one is responsible for the classification of the region according to the potential wave energy using wave height and period time series. In making the final decision on the suitability of the potential area, a large number of static land use data play an important role. These data are combined with neural network predictions for the optimizing positioning of the Wave Energy Converters. For the sake of completeness and flexibility, a Multi-Task Neural Network is developed. This model, in addition to predicting the suitability of an area depending on seagrass patterns and wave energy, also predicts land use classes through Multi-Label classification process. The proposed methodology is applied in the marine area of the city of Sines, Portugal. The first neural network achieves 98.7% Binary Classification accuracy, while the Multi-Task Neural Network 97.5% in the same metric and 93.5% in the F1 score of the Multi-Label classification output.

Humanitarian Crisis scenarios typically require immediate rescue intervention. In many cases, the conditions at a scene may be prohibitive for human rescuers to provide instant aid, because of hazardous, unexpected, and human threatening situations. These scenarios are ideal for autonomous mobile robot systems to assist in searching and even rescuing individuals. In this study, we present a synchronous ground-aerial robot collaboration approach, under which an Unmanned Aerial Vehicle (UAV) and a humanoid robot solve a Search and Rescue scenario locally, without the aid of a commonly used Global Navigation Satellite System (GNSS). Specifically, the UAV uses a combination of Simultaneous Localization and Mapping and OctoMap approaches to extract a 2.5D occupancy grid map of the unknown area in relation to the humanoid robot. The humanoid robot receives a goal position in the created map and executes a path planning algorithm in order to estimate the FootStep navigation trajectory for reaching the goal. As the humanoid robot navigates, it localizes itself in the map while using an adaptive Monte-Carlo Localization algorithm by combining local odometry data with sensor observations from the UAV. Finally, the humanoid robot performs visual human body detection while using camera data through a Darknet pre-trained neural network. The proposed robot collaboration scheme has been tested under a proof of concept setting in an exterior GNSS-denied environment.

Water vapor is one of the essential variables in monitoring the Earth’s climate. The Ocean and Land Color Instrument (OLCI) on-board the Copernicus Sentinel-3 missions measures the Integrated Water Vapor (IWV) column over land and ocean surfaces. Post-launch calibration and validation of satellite measurements constitutes a key process in the operational phase of Earth observation satellites. This work presents the external and independent validation of OLCI-A IWV product using the regional network of continuously operating Global Navigation Satellite System (GNSS) comprised 10 stations distributed over the island of Crete in the eastern Mediterranean. The Sentinel-3A/-3B OLCI imagery that captures in a single scene the entire area of Crete has been examined. For each OLCI image, the IWV value of cloud-free pixels containing the GNSS stations have been derived and compared against simultaneous GNSS-derived measurements. The absolute as well as the relative bias between OLCI-A and OLCI-B IWV measurements have been determined. There is a good agreement between OLCI and GNSS with a bias of −0.57 mm ± 2.90 mm for OLCI(A) and +2.42 ± 3.41 mm for OLCI(B). The results of this regional validation activity are compared against other studies and the regular validation carried out at the Sentinel-3 Mission Performance Center. This work concludes that the accuracy of the OLCI IWV products is within its design requirements. The potential synergy between Sentinel-2 and Sentinel-3 IWV products is also discussed.

GNSS positioning accuracy can be degraded in areas where the surrounding object geometry and morphology interacts with the GNSS signals. Specifically, urban environments pose challenges to precise GNSS positioning because of signal interference or interruptions. Also, non-GNSS surveying methods, including total stations and laser scanners, involve time consuming practices in the field and costly equipment. The present study proposes the use of an Unmanned Aerial Vehicle (UAV) for autonomous rapid mapping that resolves the problem of localization for the drone itself by acquiring location information of characteristic points on the ground in a local coordinate system using simultaneous localization and mapping (SLAM) and vision algorithms. A common UAV equipped with a camera and at least a single known point, are enough to produce a local map of the scene and to estimate the relative coordinates of pre-defined ground points along with an additional arbitrary point cloud. The resulting point cloud is readily measurable for extracting and interpreting geometric information from the area of interest. Under two novel optimization procedures performing line and plane alignment of the UAV-camera-measured point geometries, a set of experiments determines that the localization of a visual point in distances reaching 15 m from the origin, delivered a level of accuracy under 50 cm. Thus, a simple UAV with an optical sensor and a visual marker, prove quite promising and cost-effective for rapid mapping and point localization in an unknown environment.