The development of computer vision based systems dedicated to help visually impaired people to perceive the environment, to orientate and navigate has been the main research subject of many works in the recent years. A significant ensemble of resources has been employed to support the development of sensory substitution devices (SSDs) and electronic travel aids for the rehabilitation of the visually impaired. The Sound of Vision (SoV) project used a comprehensive approach to develop such an SSD, tackling all the challenging aspects that so far restrained the large scale adoption of such systems by the intended audience: Wearability, real-time operation, pervasiveness, usability, cost. This article is set to present the artificial vision based component of the SoV SSD that performs the scene reconstruction and segmentation in outdoor environments. In contrast with the indoor use case, where the system acquires depth input from a structured light camera, in outdoors SoV relies on stereo vision to detect the elements of interest and provide an audio and/or haptic representation of the environment to the user. Our stereo-based method is designed to work with wearable acquisition devices and still provide a real-time, reliable description of the scene in the context of unreliable depth input from the stereo correspondence and of the complex 6 DOF motion of the head-worn camera. We quantitatively evaluate our approach on a custom benchmarking dataset acquired with SoV cameras and provide the highlights of the usability evaluation with visually impaired users.

For most visually impaired people, simple tasks such as understanding the environment or moving safely around it represent huge challenges. The Sound of Vision system was designed as a sensory substitution device, based on computer vision techniques, that encodes any environment in a naturalistic representation through audio and haptic feedback. The present paper presents a study on the usability of this system for visually impaired people in relevant environments. The aim of the study is to assess how well the system is able to help the perception and mobility of the visually impaired participants in real life environments and circumstances. The testing scenarios were devised to allow the assessment of the added value of the Sound of Vision system compared to traditional assistive instruments, such as the white cane. Various data were collected during the tests to allow for a better evaluation of the performance: system configuration, completion times, electro-dermal activity, video footage, user feedback. With minimal training, the system could be successfully used in outdoor environments to perform various perception and mobility tasks. The benefit of the Sound of Vision device compared to the white cane was confirmed by the participants and by the evaluation results to consist in: providing early feedback about static and dynamic objects, providing feedback about elevated objects, walls, negative obstacles (e.g., holes in the ground) and signs.

Environment perception and understanding represent critical aspects in most computer vision systems and/or applications. State-of-the-art techniques to solve this vision task (e.g., semantic instance segmentation) require either dedicated hardware resources to run or a longer execution time. Generally, the main efforts were to improve the accuracy of these methods rather than make them faster. This paper presents a novel solution to speed up the semantic instance segmentation task. The solution combines two state-of-the-art methods from semantic instance segmentation and optical flow fields. To reduce the inference time, the proposed framework (i) runs the inference on every 5th frame, and (ii) for the remaining four frames, it uses the motion map computed by optical flow to warp the instance segmentation output. Using this strategy, the execution time is strongly reduced while preserving the accuracy at state-of-the-art levels. We evaluate our solution on two datasets using available benchmarks. Then, we conclude on the results obtained, highlighting the accuracy of the solution and the real-time operation capability.

Environment analysis and perception are two of the most important tasks in various application areas involving stereo vision. Specifically to autonomous driving applications these tasks are related to systems awareness and are direct linked to the decision process. This research proposes an architecture for environment analysis and perception with strait forward replicable design. This architecture is structured in three parts: input data, environment analysis and environment perception. Starting from a disparity map the architecture employs different representations that allow for ground area and obstacles detection. For free space assessment a polar grid is built allowing for better interpretation of the free space direction. In the end obstacle classification adds new information to the obstacles position and orientation. Classification is done using existing trained neural classifiers.

There are numerous methods available for evaluating leg length discrepancy (LLD), ranging from classic clinical techniques to advanced systems based on sophisticated and expensive equipment, as well as rudimentary manual adjustment mechanisms for the prosthesis by specialists. However, unilateral amputee patients often face difficulties in accessing these solutions. They either lack the necessary equipment or do not have a medical specialist available to assist them in preventing postural imbalances. This study proposes the first smartphone-based computer vision system that evaluates and automatically compensates for leg length discrepancy in transtibial prostheses, offering a low-cost, accessible, and fully autonomous alternative to existing solutions. The method was tested using complex metrological systems. The application of the proposed method demonstrated its effectiveness in correcting simulated LLD for various values. Experimental validation demonstrated the system’s ability to restore symmetry in simulated LLD cases within the 1–10 mm range, achieving a relative compensation error of 2.44%. The proposed method for correcting LLD, based on computer vision and integrated into a smartphone, represents a significant advancement in restoring symmetry for unilaterally amputated patients. This technology could provide an accessible, efficient solution, thereby reducing the need for frequent prosthetist visits and enhancing user autonomy.

Accurate estimation of grassland biomass is fundamental for designing sustainable grazing strategies and optimizing pasture management, yet conventional field methods remain labor-intensive, destructive, and difficult to scale. In this study, we exploit recent advances in computer vision to estimate multiple components of grassland biomass from overhead RGB imagery. The analysis is based on the Image2Biomass dataset, comprising 1162 annotated images of grasslands across Australia, each paired with laboratory-validated biomass measurements. A structured preprocessing pipeline was implemented, including exploratory data analysis, outlier mitigation, and logarithmic transformation of target variables, in accordance with the dataset evaluation protocol. Building on this foundation, we propose an encoder–decoder regression framework that integrates self-supervised visual representation learning with ensemble-based prediction. The encoder employs a DINOv2 Giant model as a feature extractor to capture detailed spatial and structural characteristics of the sward, while the decoder uses a stacking ensemble combining LightGBM, XGBoost, and Ridge Regression. Across 15 repetitions of shuffled four-fold cross-validation, the cross-fitted stacking ensemble achieved a weighted coefficient of determination of $R_w^2=0.7757\pm 0.0171$, a weighted mean absolute error of $8.3867\pm 0.2736$ g, and a weighted root mean squared error of $13.3994\pm 0.4995$ g. The ensemble significantly outperformed LightGBM, XGBoost, and Ridge Regression on the primary weighted $R^2$ metric in paired comparisons (Holm-adjusted $p<0.001$ for all three comparisons). These results highlight the potential of computer vision methods as scalable, non-destructive tools for operational monitoring of grassland biomass, supporting more informed agronomic decision-making in pasture-based systems.