Professional swimming coaches make use of videos to evaluate their athletes’ performances. Specifically, the videos are manually analyzed in order to observe the movements of all parts of the swimmer’s body during the exercise and to give indications for improving swimming technique. This operation is time-consuming, laborious and error prone. In recent years, alternative technologies have been introduced in the literature, but they still have severe limitations that make their correct and effective use impossible. In fact, the currently available techniques based on image analysis only apply to certain swimming styles; moreover, they are strongly influenced by disturbing elements (i.e., the presence of bubbles, splashes and reflections), resulting in poor measurement accuracy. The use of wearable sensors (accelerometers or photoplethysmographic sensors) or optical markers, although they can guarantee high reliability and accuracy, disturb the performance of the athletes, who tend to dislike these solutions. In this work we introduce swimmerNET, a new marker-less 2D swimmer pose estimation approach based on the combined use of computer vision algorithms and fully convolutional neural networks. By using a single 8 Mpixel wide-angle camera, the proposed system is able to estimate the pose of a swimmer during exercise while guaranteeing adequate measurement accuracy. The method has been successfully tested on several athletes (i.e., different physical characteristics and different swimming technique), obtaining an average error and a standard deviation (worst case scenario for the dataset analyzed) of approximately 1 mm and 10 mm, respectively.

This study presents a comprehensive methodology for the detection and characterization of fractures in geological samples using X-ray computed tomography (CT). By combining convolution-based image processing techniques with advanced neural network-based segmentation, the proposed approach achieves high precision in identifying complex fracture networks. The method was applied to a marly limestone sample from the Maiolica Formation, part of the Umbria–Marche stratigraphic succession (Northern Apennines, Italy), a geological context where fractures often vary in size and contrast and are frequently filled with minerals such as calcite or clays, making their detection challenging. A critical part of the work involved addressing multiple sources of uncertainty that can impact fracture identification and measurement. These included the inherent spatial resolution limit of the CT system (voxel size of 70.69 μm), low contrast between fractures and the surrounding matrix, artifacts introduced by the tomographic reconstruction process (specifically the Radon transform), and noise from both the imaging system and environmental factors. To mitigate these challenges, we employed a series of preprocessing steps such as Gaussian and median filtering to enhance image quality and reduce noise, scanning from multiple angles to improve data redundancy, and intensity normalization to compensate for shading artifacts. The neural network segmentation demonstrated superior capability in distinguishing fractures filled with various materials from the host rock, overcoming the limitations observed in traditional convolution-based methods. Overall, this integrated workflow significantly improves the reliability and accuracy of fracture quantification in CT data, providing a robust and reproducible framework for the analysis of discontinuities in heterogeneous and complex geological materials.

The acoustic analysis of a moving object, such as in pass-by or fly-over tests, is a very important and demanding issue. These types of analyses make it possible to characterize the machine in quite realistic conditions, but the typical difficulties related to source localization and characterization are usually exacerbated by the need to take into consideration and to compensate for the object movement. In this paper, a technique based on acoustic beamforming is proposed, which is applicable to all those cases where the object under investigation is moving. In the proposed technique, the object’s movement is not regarded as a problem but as a resource, enabling a virtual increase in the number of microphone acquisitions. For a stationary acoustic emission from a moving object, each time segment of the acquired signal is treated as if it is coming from a microphone (virtual) positioned differently relative to the object’s reference system. This paper describes the technique and presents examples of results obtained from both simulated and real signals. Performance analysis is conducted and discussed in detail.

This study presents and validates two simplified methods to estimate the frontal area of swimmers during active motion, based on video recordings from frontal and lateral views. The aim is to provide low-complexity and practical tools suitable for use in some everyday sport training environments, such as swimming pools. The frontal view enables a direct measurement of the true projected area, but is affected by factors such as the water transparency and the variable swimmer-camera distance. Although providing only a 2D projection, the lateral view benefits from a short working distance and good stability, improving the analysis of posture-related features such as body alignment and angle of incidence. By tracking and analyzing the evolution of the frontal area over the stroke cycle, the proposed methods allow the separation of propulsive and resistive contributions, revealing stroke-specific technical patterns. The results show that this approach can detect changes in technique, posture, and performance across different swimming styles and for the same athlete under varying conditions. Despite limitations in absolute accuracy due to simplified calibration and environmental variability, the system demonstrates sufficient repeatability and sensitivity for comparative analysis. The methods support biomechanical evaluation and feedback, contributing to technique refinement and improved swimmer performance.