This paper describes a Mission Definition System and the automated flight process it enables to implement measurement plans for discrete infrastructure inspections using aerial platforms, and specifically multi-rotor drones. The mission definition aims at improving planning efficiency with respect to state-of-the-art waypoint-based techniques, using high-level mission definition primitives and linking them with realistic flight models to simulate the inspection in advance. It also provides flight scripts and measurement plans which can be executed by commercial drones. Its user interfaces facilitate mission definition, pre-flight 3D synthetic mission visualisation and flight evaluation. Results are delivered for a set of representative infrastructure inspection flights, showing the accuracy of the flight prediction tools in actual operations using automated flight control.

Advancements in embedded systems and Artificial Intelligence (AI) have enhanced the capabilities of Unmanned Aircraft Vehicles (UAVs) in computer vision. However, the integration of AI techniques o-nboard drones is constrained by their processing capabilities. In this sense, this study evaluates the deployment of object detection models (YOLOv8n and YOLOv8s) on both resource-constrained edge devices and cloud environments. The objective is to carry out a comparative performance analysis using a representative real-time UAV image processing pipeline. Specifically, the NVIDIA Jetson Orin Nano, Orin NX, and Raspberry Pi 5 (RPI5) devices have been tested to measure their detection accuracy, inference speed, and energy consumption, and the effects of post-training quantization (PTQ). The results show that YOLOv8n surpasses YOLOv8s in its inference speed, achieving 52 FPS on the Jetson Orin NX and 65 fps with INT8 quantization. Conversely, the RPI5 failed to satisfy the real-time processing needs in spite of its suitability for low-energy consumption applications. An analysis of both the cloud-based and edge-based end-to-end processing times showed that increased communication latencies hindered real-time applications, revealing trade-offs between edge (low latency) and cloud processing (quick processing). Overall, these findings contribute to providing recommendations and optimization strategies for the deployment of AI models on UAVs.

Noncollaborative surveillance of airborne UAS (Unmanned Aerial System) is a key enabler to the safe integration of UAS within a UTM (Unmanned Traffic Management) ecosystem. Thus, a wide variety of new sensors (known as Counter-UAS sensors) are being developed to provide real-time UAS tracking, ranging from radar, RF analysis and image-based detection to even sound-based sensors. This paper aims to discuss the current state-of-the art technology in this wide variety of sensors (both academically and commercially) and to propose a set of simulation models for them. Thus, the review is focused on identifying the key parameters and processes that allow modeling their performance and operation, which reflect the variety of measurement processes. The resulting simulation models are designed to help evaluate how sensors’ performances affect UTM systems, and specifically the implications in their tracking and tactical services (i.e., tactical conflicts with uncontrolled drones). The simulation models cover probabilistic detection (i.e., false alarms and probability of detection) and measurement errors, considering equipment installation (i.e., monostatic vs. multistatic configurations, passive sensing, etc.). The models were integrated in a UTM simulation platform and simulation results are included in the paper for active radars, passive radars, and acoustic sensors.

Unmanned traffic management (UTM) systems rely on collaborative position reporting to track unmanned aerial system (UAS) operations over wide unsurveilled (with counter-UAS systems) areas. Many different technologies, such as Remote-ID, ADS-B, FLARM, or MLAT might be used for this purpose, in addition to the direct exploitation of C2 telemetry, relayed though cellular networks. This paper provides an overview of the most used collaborative sensors and surveillance systems in this context, analyzing their main technical parameters and performance effects. In addition, this paper proposes an abstracted general statistical simulation model covering message encoding, network capacity and access, sensors coverage and distribution, message transmission and decoding. Making use of this abstracted model, this paper proposes a particularized set of simulation models for ADS-B, FLARM and Remote-Id; it is thus useful to test their potential integration in UTM systems. Finally, a comparative analysis, based on simulation, of these systems, is performed. It is shown that the most relevant effects are those related with quantification and the potential saturation of the communication channels leading to collisions and delays.

Unmanned traffic management (UTM) systems will become a key enabler to the future drone market ecosystem, enabling the safe concurrent operation of both manned and unmanned aircrafts. Currently, these systems are usually tested by performing real scenarios that are costly, limited, hardly scalable, and poorly repeatable. As a solution, in this paper we propose an agent-based simulation platform, implemented through a micro service architecture, which may simulate UTM information sources, such as flight plans, telemetry messages, or tracks from a surveillance network. The final objective of this simulator is to use these information streams to perform a system-level evaluation of UTM systems both in the pre-flight and in-flight stages. The proposed platform, with a focus on simulation of communications and sensors, allows to model UTM actors’ behaviors and their interactions. In addition, it also considers the manual definition of events to simulate unexpected behaviors/events (contingencies), such as communications failures or pilots’ actions. In order to validate our architecture, we implemented a simulator that considers the following actors: drones, pilots, ground control stations, surveillance networks, and communications networks. This platform enables the simulation of the drone trajectory and control, the C2 (command and control) link, drone detection by surveillance sensors, and the communication of all agents by means of a mobile communications network. Our results show that it is possible to truthfully recreate complex scenarios using this simulator, mitigating the disadvantages of real testbeds.

Accurate human posture classification in images and videos is crucial for automated applications across various fields, including work safety, physical rehabilitation, sports training, or daily assisted living. Recently, multimodal learning methods, such as Contrastive Language-Image Pretraining (CLIP), have advanced significantly in jointly understanding images and text. This study aims to assess the effectiveness of CLIP in classifying human postures, focusing on its application in yoga. Despite the initial limitations of the zero-shot approach, applying transfer learning on 15,301 images (real and synthetic) with 82 classes has shown promising results. The article describes the full procedure for fine-tuning, including the choice for image description syntax, models and hyperparameters adjustment. The fine-tuned CLIP model, tested on 3826 images, achieves an accuracy of over 85%, surpassing the current state-of-the-art of previous works on the same dataset by approximately 6%, its training time being 3.5 times lower than what is needed to fine-tune a YOLOv8-based model. For more application-oriented scenarios, with smaller datasets of six postures each, containing 1301 and 401 training images, the fine-tuned models attain an accuracy of 98.8% and 99.1%, respectively. Furthermore, our experiments indicate that training with as few as 20 images per pose can yield around 90% accuracy in a six-class dataset. This study demonstrates that this multimodal technique can be effectively used for yoga pose classification, and possibly for human posture classification, in general. Additionally, CLIP inference time (around 7 ms) supports that the model can be integrated into automated systems for posture evaluation, e.g., for developing a real-time personal yoga assistant for performance assessment.