Ego-Centric Vision for Hand Occupancy Detection: Toward Safer Human-Robot Interaction
Loading...
Date
2026
Authors
Матійчик, Тимофій
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The central subject of Human-Robot Interactions studies are robots, that are supposed to collaborate with humans, thus operate in a common humans’ workspace and understand their intents. One of the most important domains concerns in modern HRI researches is interactions safety – how robot should collaborate with humans and provide physically and socially secure interactions. Human hands occupancy in that case are one of key evidences for robots to define safe and appropriate interactions. The objective of this work is to reveal the most effective and safe solution, based on Computer Vision techniques, for hands segmentation and occupancy detection for human-robot interactions. The solution must demonstrate the best correlation of detection performance and effectiveness on robotic system’s integration that will provide the most reliable behavior in its interactions or collaboration with human. To accomplish the objective, the study explores the capabilities of 3 main approaches of hands segmentation and occupancy classification: a classical technique that relies on classical CV algorithms and heuristic estimations; the deep learning that applies deep segmentation and vision-language models; the hybrid method that combines classical segmentation and deep learning classification. These techniques are evaluated in terms of segmentation and classification performance and technical efficiency. This evaluation demonstrates both predictive capabilities and practical suitability and provides insights for defining the best approach for enhancing safety in HRI and adoption to robotic systems.
Description
Keywords
Human-Robot Interaction (HRI), Computer Vision, Deep Learning, Vision-Language Models (VLM), hands segmentation, bachelor`s thesis