School of Engineering Department of Computer Science and Engineering 90 Video Analytics and IoT People/Asset Sensing for Smart City Applications Supervisor: CHAN Gary Shueng Han / CSE Student: KAO Ka Ho / CPEG Course: UROP 1100, Spring This project aims to develop a system that assists users in accurately performing sports actions by providing feedback and scoring. Users will first follow the sport action performed in the given video. The system will then provide them a score based on their performance. It will employ Dynamic Time Warping (DTW) to compare the user’s motions with the reference action by extracting and comparing user key points from the video. Individual scores for each body part, like the hands, legs, and head, will also be available, allowing users to identify their problems more accurately. This helps users improve their actions effectively by reviewing their scores and mistakes. Video Analytics and IoT People/Asset Sensing for Smart City Applications Supervisor: CHAN Gary Shueng Han / CSE Student: WU Minghua / COMP Course: UROP 1100, Fall UROP 2100, Spring This semester, I participated in the discussion about the application of GRAFICS with E-LINE algorithm in M+ and Xiqu Center. This project aims to help visitors locate their floors on their phones by using pure WiFi signals. There are several challenges in this project. There is an atrium inside the M+ which leads to the situation that users receive strong signals from other floors. The GRAFICS with E-LINE algorithm is used in this project to check if this model can provide a good result under this complex situation. Several modifications are made to try out different design decisions in the online inference phase. The current result is still not as good as expected and the project is still going on. AI in Medical Imaging: Automatic Stroke Analysis on Brain CT Scans Supervisor: CHEN Hao / CSE Student: LEE Hsin-ning / COMP Course: UROP 2100, Fall Ischemic stroke (IS) is a leading cause of death and long-term disability globally, resulting from impaired blood flow to the brain. CT perfusion (CTP) is a widely used diagnostic method for IS, involving the acquisition of 3D X-ray images. As it is difficult to directly interpret raw CT images, perfusion parameter maps derived from the raw CT scans are required to precisely indicate the problematic areas. Currently, clinical practice relies on deriving these maps from CTP scans using Arterial Input Function (AIF). Nevertheless, raw CTP scans frequently suffer from sensitivity issues, noise, and artifacts, leading to less reliable conversions. Moreover, the selection of AIF can impact the accuracy of the parameter maps. To address these challenges, we utilize of Deep Learning-driven methods for stroke lesion segmentation and parameter map estimation. This approach not only expedites the treatment process but also enhances diagnostic accuracy. Our project introduces a semi-supervised framework that combines parameter map prediction and ischemic lesion segmentation within a shared model. To enhance the efficiency of the model, we conducted various experiments based on recent advancements. Considering low temporal resolutions, we explore strategies to effectively utilize limited data, aiming to achieve satisfactory results.
RkJQdWJsaXNoZXIy NDk5Njg=