School of Engineering Department of Computer Science and Engineering 97 AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: CHI Yankuan / COSC Course: UROP 4100, Fall 3D visual localization is a famous task in computer vision. There are several categories of methods for tackling this task. One main-stream category is feature-matching methods, which calculate the pose of a test image by matching 2D and 3D features to obtain 2D-3D correspondences. In this project, I explored featurematching localization models. The pipeline is mostly based on the PNeRFLoc. Then I tried to develop a fine matching module based on the current proposed models. Moreover, I found that adding a reliability prediction module on top of the model can produce better results. Also, I conducted a large number of experiments to verify the effect of the above methods. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: FU Suen Man / DSCT Course: UROP 1000, Summer This report provides a detailed presentation, encompassing description, methodology, and statistical findings, of three geospatial visualizations depicting wireless coverage patterns, demonstrating the distribution and signal propagation characteristics of IoT devices throughout Hong Kong districts. These geospatial visualizations can facilitate the research project by providing spatially explicit representations of wireless signal coverage and IoT device distribution patterns across Hong Kong’s heterogeneous urban landscape. By translating complex signal propagation metrics and device location data into intuitive visual formats—including density maps, signal strength heatmaps, and dynamic agent-based trajectories—the visualizations enable researchers to identify spatial correlations, detect coverage gaps, and analyze usage trends, that can facilitate the extraction of actionable insights for user analytics and personalized recommendations. AI meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: KANG Zhaowei / COSC Course: UROP 4100, Fall Radio frequency signals play a crucial role in various applications, yet their signal fingerprint collection remains labor-intensive and time-consuming. To overcome this challenge, we present a multi-site fingerprint augmentation model designed to generate high-density fingerprint data based on the sparse signal fingerprint, thus greatly reducing the labor force needed. Our model enhances signal density by transforming sparse signals into dense signals through the integration of both patch-level and site-level models. It is also capable of synthesizing fingerprint signals for different sites. We also validated the effectiveness and robustness of our methodology using several real-world datasets to show that Our approach significantly improves the density of the fingerprint signal.
RkJQdWJsaXNoZXIy NDk5Njg=