Aided in the development of a research paper, where I developed a Java / React tool which automatically converts a dataset into Web Ontology Language (OWL) format.
The tool allows users to manually insert / copy a dataset into a dynamic table provided by the GUI and produces a self-contained OWL ontology representing the target dataset.
My capstone project was a combined effort with students from both UTS as well as Hong Kong Polytechnic University (PolyU). Each student investigated the effectiveness of deep learning in the development of Advanced Driver Assistance Systems (ADAS), exploring topics such as: Pedestrian detection; Vehicle detection; Traffic-sign recognition; Lane detection; and Drowsy Driver detection. The finished implementations for each topic were then combined to create a complete ADAS system.
Additionally, in order to prepare and synthesise our knowledge and ideas for the main project, the capstone team participated in the IEEE Video and Image Processing Cup (VIP Cup 2019). The challenge focused on the ability to create a deep learning model that would be able to detect and recognise a total of 18 human activities from an FPV chest-mounted camera, which included activities such as walking; chatting; printing; reading; and using a mobile phone. Our model was also required to be able to conceal (blur) private information such as faces, or the content of a computer screen.
Our capstone team proposed a new Spatio-temporal attention network (STANet), which focuses on analysing both a combination of sRGB frames, as well as each frame’s optical flows, through the use of ResNet50, ResNet34, and LSTM modules.
We attended and presented this model at the finals, hosted at the IEEE International Conference on Image Processing (ICIP 2019) in Taipei (Taiwan), where we achieved 1st Place.