School of Engineering Department of Computer Science and Engineering 100 Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: TANG Kewei / COMP Course: UROP 1100, Spring This report delves into the foundational concepts and existing research pertinent to indoor localization, with a specific focus on the technologies: Inertial Measurement Unit (IMU)-based localization and 3D reconstruction. It begins with a broad overview of indoor IMU-based localization techniques on Pedestrian Dead Reckoning (PDR) and deep learning enhancements for IMU data such as Ronin. Then, it discusses traditional and advanced 3D reconstruction methods including Colmap, Dust3R, and Fast3R, and finally, explores synergistic approaches that combine these modalities for enhanced performance. Video Analytics and IoT People/Asset Sensing for Smart City Applications Supervisor: CHAN Gary Shueng Han / CSE Student: WU Minghua / COMP Course: UROP 3100, Fall UROP 4100, Spring This semester, I participated in the application of the FIS-ONE model in real-world scenarios. I continue to focus on migrating the PyTorch version of FIS-ONE to TensorFlow so that it can be used in the TensorFlow Lite environment and the inference can be done in edge devices. Regarding the progress, based on the work in the last semester, the model construction in TensorFlow is done. The model is exported and a sample TFLite application based on C++ is written so that the model can be used for inference in the edge devices. The integration in the Android environment is still in progress. In this report, I will share the understanding of FIS-ONE, the model construction in TensorFlow, and the implementation in the TFLite C++ environment for edge inference. Ambient Intelligence Empowered Smart Nursing Home for Vision-Based Elderly Caring Supervisor: CHEN Hao / CSE Student: CHEN Yiyang / CPEG Course: UROP 1100, Summer This report summarizes my summer research experience on the project Ambient Intelligence Empowered Smart Nursing Home for Vision-Based Elderly Caring. I initially contributed to a large-scale human evaluation study, where our group conducted video annotation, resulting in frame-level labels for over 7,000 video clips, captured over a continuous 21-day period, which is the largest dataset for human evaluation in vision-based ambient intelligence up to now. Motivated to deepen my understanding, I focused on facial expression recognition and replicated the code of the paper “Estimation of Continuous Valence and Arousal Levels from Faces in Naturalistic Conditions”, learning computer vision and deep learning fundamentals along the way. I later implemented simplified ResNet-18 architecture and trained it on the MNIST dataset, achieving 92.25% accuracy. This experience boosts my interest in computer vision and deep learning.
RkJQdWJsaXNoZXIy NDk5Njg=