Scanpath analysis provides a powerful window into visual behavior by jointly capturing the spatial organization and temporal dynamics of gaze. By linking perception, cognition, and oculomotor control, scanpaths offer rich insights into how individuals explore visual scenes and accomplish task goals. Despite decades of research, however, the field remains methodologically fragmented, with a wide diversity of representations and comparison metrics that complicate interpretation and methodological choice. This article reviews computational approaches for the characterization and comparison of scanpaths, with an explicit focus on their underlying assumptions, interpretability, and practical implications. We first survey representations and metrics designed to describe individual scanpaths, ranging from geometric descriptors and spatial density representations to more advanced approaches such as attention maps, recurrence quantification analysis, and symbolic string encodings that capture temporal regularities and structural patterns. We then review methods for comparing scanpaths across observers, stimuli, or tasks, including point-mapping metrics, elastic alignment techniques, string-edit distances, saliency-based measures, and hybrid approaches integrating spatial and temporal information. Across these methods, we highlight their respective strengths, limitations, and sensitivities to design choices such as discretization, spatial resolution, and temporal weighting. Rather than promoting a single optimal metric, this review emphasizes scanpath analysis as a family of complementary tools whose relevance depends on the research question and experimental context. Overall, this work aims to provide a unified conceptual framework to guide methodological selection, foster reproducibility, and support the meaningful interpretation of gaze dynamics across disciplines.
Numerical simulations are crucial for modeling complex systems, but calibrating them becomes challenging when data are noisy or incomplete and likelihood evaluations are computationally expensive. Bayesian calibration offers an interesting way to handle uncertainty, yet computing the posterior distribution remains a major challenge under such conditions. To address this, we propose a sequential surrogate-based approach that incrementally improves the approximation of the log-likelihood using Gaussian Process Regression. Starting from limited evaluations, the surrogate and its gradient are refined step by step. At each iteration, new evaluations of the expensive likelihood are added only at informative locations, that is to say where the surrogate is most uncertain and where the potential impact on the posterior is greatest. The surrogate is then coupled with the Metropolis-Adjusted Langevin Algorithm, which uses gradient information to efficiently explore the posterior. This approach accelerates convergence, handles relatively high-dimensional settings, and keeps computational costs low. We demonstrate its effectiveness on both a synthetic benchmark and an industrial application involving the calibration of high-speed train parameters from incomplete sensor data.
Our study investigates how high-speed transport affects humans' perception of time, space, and speed. We tested 247 passengers traveling on the French Paris-Lyon high-speed train (TGV) who estimated the traveled distance, elapsed duration, and instantaneous and average speeds, while rating their confidence, felt speed of time passage, and emotional state. Environmental variables in this In Situ design included seating orientation (forward or backward) and train acceleration (positive, null [constant high-speed], or negative [deceleration]). Our findings show that participants' magnitude (duration, distance) estimations scaled with magnitudes. Estimations also exhibited typical features such as central tendency, which interestingly scaled to the midpoint of the journey (∼69 min) rather than the expected ∼15 min. Surprisingly, we found no influence of emotions or felt speed of time passage on magnitude and speed estimates. Instead, participants' prior knowledge of the journey's total duration and distance better explained their magnitude estimations than their on-the-fly speed estimates. Seating orientation did not significantly affect estimations, whereas acceleration partly did. Participants demonstrated reliable metacognitive ratings of their duration and instantaneous speed estimates, but not of their distance and average speed estimates. Altogether, our results show that in passive high-speed transport, passengers may rely on cognitive heuristics (mental maps) rather than sensorimotor integration (path integration) to estimate magnitudes. We discuss the implications of our work for understanding mental representations of magnitudes and speed in real-life situations.
Railway geometry monitoring is essential for track maintenance. Recent developments have led to the use of accelerometers mounted on the axle-boxes of commercial trains to measure vertical rail irregularities. The aim of the work presented hereunder is twofold. First, the uncertainties involved in estimating the vertical offset and cross-level of railway tracks by double integration of accelerations measured in axle-boxes are quantified and discussed using simulation. This first part of the work is carried out with the help of railway dynamics simulation, which enables the environmental conditions, system inputs, and train reaction to be perfectly controlled. Particular attention is paid to understanding the displacements and rotations of the sensors placed in the axle-boxes. Second, based on these elements, an enhanced reconstruction that is more accurate, particularly for long wavelengths, is proposed. Finally, an application to measured signals is presented.
