Work

Generació de dades sintètiques d'IMU a partir de l'estimació de la posició 3D basada en vídeo per al reconeixement d'activitats humanes

Year: 2025
Type: dissertation
Abstract: [EN] In recent years, Human Activity Recognition (HAR) has garned increasing interest, driven by its applications in healthcare, sports, smart environments, and humancomputer interaction. One of the main challenges in the development of models to deal with such task is the availability and quality of data to train and validate such models. This final project will address this issue by focusing in the development of a model capable of obtaining synthetic Inertial Measurement Unit (IMU) data from 3D pose data extracted from 2D video. The tool will enable researchers and developers to simulate realistic signals of a body-placed IMU sensor using only in-the-wild physical activity videos. The goal is for this approach is to contribute to the robustness and scalability of HAR models. It is expected that the developed model will facilitate the training of the model by using data augmentation, reduce dependency on costly sensor-based data collection, and support the development of more accurate and adaptable HAR systems.
Source: Hispana
Language: Spanish
Open Access status: green
OpenAlex ID: W7206667592
Cites:
Cited by:
Related to: 10