Human activity recognition (HAR) is an area of research that has received increasing attention in recent years. HAR can be applied in many areas, such as human computer interaction (HCI), smart homes, internet of things (IoT), detection of abnormal or suspicious behaviours, fitness trackers and much more. Millimetre wave radar (mmWave) is an up-and-coming development in this area, as it has several advantages over other technologies. Unlike optical sensors, mmWave is not reliant on external lighting conditions and has vastly reduced privacy concerns.
This research explores alternative methods of activity segmentation to allow for the detection of activity ends. This is in an effort to improve the recognition of complex activity sequences, which consist of multiple activities in succession, and may constitute a higher level behaviour. Existing research fails to cover the recognition of longer, more complex activity sequences and assumes activities fit into a window of a fixed length. We propose a novel activity recognition method using millimetre wave (mmWave) radar, one that selects salient activity data frames, and allows for the recognition of activity sequences using machine learning.
Full organized source files for the project.
This is the final location of our source code.
Reports, references and components such as diagrams.
This is the location of the Compendium Report, and any other required documentation. This folder also contains the source files for our poster, and any diagrams created with photoshop.
Files and code related to the testing and development of the project.
These files are relatively unorganized and not designed for examination, but have been included as a reference for our entire development process.
Other areas of work, such as the radar mount.