Scientific Reseach

Our co-founder has conducted extensive scientific research over the years, contributing significantly to the fields of computer vision, human recognition, and health monitoring through advanced technological methods. These studies have led to the development of innovative solutions in posture detection, fatigue monitoring, and 3D object recognition. Below is a list of key research publications authored by our co-founder:

📜A Human-Adaptive Model for User Performance and Fatigue Evaluation during Gaze-Tracking Tasks

Link: https://www.mdpi.com/2079-9292/12/5/1130

Proposes a human-adaptive model to evaluate user performance and fatigue during gaze-tracking tasks. Using a deep recurrent hierarchical network (DRHN) model, it incorporates biofeedback to enhance human-computer interaction (HCI). The model's validity was tested with 12 volunteers playing a gaze-controlled game, demonstrating its effectiveness in monitoring user performance dynamics, fatigue, and recovery. This adaptive approach aims to improve the usability of physiological-computing-based user interfaces, particularly in applications like assisted living, healthcare, and rehabilitation.

📜3D Object Reconstruction from Imperfect Depth Data Using Extended YOLOv3 Network

Link: https://www.mdpi.com/1424-8220/20/7/2025

This paper discusses an advanced method for reconstructing 3D objects using incomplete depth data. It proposes a hybrid neural network model that improves reconstruction accuracy by 8.53%, effectively filling occluded areas and reducing noise. The method uses segmentation masks and instance classification, making it suitable for general scene reconstruction, not just single objects. This approach addresses challenges like limited data availability and occlusions, offering significant improvements over traditional techniques.

📜Human Recognition Through a Webcam Using Deep Learning Techniques

Link: https://ieeexplore.ieee.org/document/9594811

Explores the use of advanced deep learning methods to accurately identify and recognize human actions and behaviors through webcam footage. It proposes a robust framework that leverages convolutional neural networks (CNNs) and other machine-learning models to enhance the precision and efficiency of human recognition. This approach aims to improve applications in security, healthcare, and human-computer interaction by providing reliable and real-time human activity monitoring. The study highlights significant advancements in accuracy and processing speed compared to traditional recognition methods.

📜Detection of Sitting Posture Using Hierarchical Image Composition and Deep Learning

Link: https://peerj.com/articles/cs-442/

This paper presents a novel method for detecting sitting posture using a deep recurrent hierarchical network (DRHN) model based on MobileNetV2. The approach addresses issues like occlusion and visibility of the human torso in a frame by utilizing RGB-Depth frame sequences. It achieves 91.47% accuracy at a 10 fps rate, making it suitable for applications in assisted living, healthcare, and rehabilitation by accurately recognizing and classifying sitting postures to improve ergonomics and reduce health risks associated with poor posture.