Human Activity Recognition (HAR)
Human activity recognition (HAR) is one of the most popular research topics in the world. With the recent development of sensor technologies and machine learning techniques, many HAR systems have been proposed and developed. Such systems often use one of cameras, sensors, and wearable devices or a mixture of them to analyse human behaviour. Camera-based methods have been shown to achieve outstanding performance for various tasks, from gesture recognition to posture recognition. However, cameras are intrusive, and many people would be concerned about privacy. Various types of sensors, like radio frequency (RF) signal transceivers and environmental sensors, are also a popular choice for HAR. Sensors monitor and detect changes in the environment caused by human subjects, where the information can be analysed to recognize the corresponding human activity. Researchers have proposed the use of ultrasonic sensors for gait estimation and WiFi sensors for human localizing. Pictorial representation of the HAR is given in the figure.
Courtesy: dimsstudio.org
Most of these techniques are designed for particular use cases and only work under certain environments. Investigates the potential use of millimetre-wave (mmWave) radars for HAR. The mmWave radars we selected from TI operate at 76 to 81 GHz and have a maximum of 4 GHz available bandwidth. With FMCW techniques, the high bandwidth allows object detection at a high resolution of around 4 cm, mmWave radars are nonintrusive and are able to sense in various conditions including darkness, smoke, and fog, which are crucial in many applications. Although mmWave radars do not provide dense information as a camera would, they provide high-resolution in distance, velocity, and angle estimation of the objects in the scene, which can be potentially very useful for understanding their status and motion, as well as distinguishing the object of interest from background clutter.
Dr. Mohd Israil
Associate Professor
Physics Department, Faculty of Science