About the Project
Human Activity Recognition (HAR) has many applications from sport tracking, and healthcare, to robotics and automatous vehicles. It is often implemented using cameras or wearable sensors. However, these have several drawbacks; wearable sensors have to be placed on the body, and cameras rely on line of sight vision and good illumination. An alternative method is to analyse the stray reflected signals that are generated using common Wi-Fi signals. As a person walks through the EM field, these reflections can be measured. This can cope with occlusions, for example seeing through a wall.
By training deep learning models, we can then infer what a person is actually doing. This method is not however perfect. Issues such as the number of people that can be tracked at once is a limitation, as is noise and the nature and amount of occlusions. Increasing the coverage, better models and combining the passive Wi-Fi and computer vision together could potentially produce more accurate results when presented with complex environments. This PhD will explore this research area.
Potential applications include human activity recognition, tracking people, tracking crowds, medical movement disorder diagnosis (e.g. Parkinson’s disease), sports and healthcare analysis.