Abstract
Human activity recognition (HAR) has been studied for decades in computer vision and has shown great success. However, as people are caring more about privacy issues, researchers are investigating non-intrusive HAR systems using radar-based techniques, among which millimetre-wave (mmWave) radars have received great popularity due to their capability of capturing high-resolution spatial information about the scene. This thesis presents a systematic study of HAR using mmWave radars. It explains the fundamentals of mmWave sensing techniques, discusses its use in HAR applications, and highlights the challenge of the sparse and noisy data through a purpose-built simulation system that can import arbitrary 3D models to form a scene and simulate the radar signal with configuration antenna settings. A software framework for managing multiple radars is presented that allows real-time data transmission, data processing, and result visualization.Based on the software framework, three HAR systems are presented. First, a human detection and tracking system is presented as the fundamental of HAR. The system operates two radars simultaneously that verify each other’s detection and significantly reduce the probability of false alarms. The system achieves 90.4% sensitivity and 98.6% precision when detecting up to four people in the room. Then, a human posture estimation system is presented that uses two radars as a vertical array and a neural network model to estimate the joint positions of the person. The system achieved over 71.3% accuracy when detecting postures that are commonly seen in an office environment with arbitrary limb motions. Finally, a human vital sign detection system is presented that uses one mmWave radar to detect a person’s heart rate when exercising on a treadmill. It overcomes the challenge that the heartbeat signal can be difficult to extract when there is body movement, and achieved a low error rate of 5.4%.
| Date of Award | 27 Sept 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Naim Dahnoun (Supervisor) & Krishna Coimbatore Balram (Supervisor) |