Study of hand gesture recognition using impulse radio ultra wideband (IRUWB) radar sensor

• Main Author: Terence Jerome Daim
• Format: Thesis
• Language: English; English
• Published: 2023
• Subjects: TK5101-6720 Telecommunication Including telegraphy, telephone, radio, radar, television
• Online Access: https://eprints.ums.edu.my/id/eprint/39094/1/24%20PAGES.pdf; https://eprints.ums.edu.my/id/eprint/39094/2/FULLTEXT.pdf; https://eprints.ums.edu.my/id/eprint/39094/
• Institution: Universiti Malaysia Sabah

Hand gesture recognition technology has gained significant attention in recent years due to its potential to revolutionize human-computer interaction by offering a natural and intuitive means of communication. This work addresses the limitations of existing systems and focuses on developing a novel hand gesture recognition system that leverages Impulse Radio Ultra-Wide Band (IR-UWB) radar sensors. The primary objective of this work is to create a comprehensive hand gesture recognition system capable of accurately recognizing a wide range of hand gestures while distinguishing between them based on gesture speed. To achieve this, this work defines three key objectives. First objective is to determine the optimal setup for IR-UWB radar sensor data acquisition, considering factors such as sensor placement and configuration. Second objective is to develop and assess hand gesture recognition models using seven different classifiers to achieve accurate and reliable recognition of hand gestures. Third objective is to analyse the performance of the developed classifiers in comparison to existing research in the field, with a focus on recognizing both hand gestures and their associated speeds. The work begins by providing insights into the state of the art in hand gesture recognition and IR-UWB radar sensor technology. Data collection experiments yield a diverse dataset of hand gestures, including variations in speed, essential for algorithm development. The developed algorithms interpret raw IR-UWB radar sensor data and associate it with specific hand gestures, addressing the core objective of gesture recognition. Speed recognition integration further enhances the system's ability to distinguish between gestures performed at different speeds. The resulting hand gesture recognition system is rigorously evaluated and compared to existing methods, demonstrating its effectiveness. Documentation of the development process ensures the methodology and findings are well-documented for reference and replication. While this research makes significant contributions to the field of hand gesture recognition, it also identifies several areas for future work. These include exploring recognition of gestures performed by two hands simultaneously, scalability to different environments, optimal sensor placement, and addressing user variability. Seven classification algorithms (K-Nearest Neighbour, Logistic Regression, Naive Bayes, Gradient Boosting, AdaBoost, Bagging, and Linear Discriminant Analysis) were meticulously explored for hand gesture recognition. The evaluation, based on macro F1 scores to balance precision and recall, aimed to assess their effectiveness. Linear Discriminant Analysis proved most accurate, especially in fast hand gestures, emphasizing its significance in real-time applications. In contrast, AdaBoost exhibited weaker performance, indicating areas for improvement. A slight accuracy decrease for "Up-Down" and "Down-Up" gestures compared to existing literature. However, it significantly outperforms certain literature by 16.28% for "Left-Right" gestures at slow speeds, showcasing improved recognition and robustness. Additionally, the research enhances system functionality, enabling intricate interactions. A developed application allows users to visualize executed hand gestures, paving the way for future integration of complex interaction sub-systems in various gesture recognition applications. In summary, this work advances the field of hand gesture recognition by introducing a novel IR-UWB radar-based system that accurately recognizes hand gestures and distinguishes their speeds, offering improved performance and usability for a wide range of applications.