School of Engineering Department of Computer Science and Engineering 103 Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: HU Chenxi / COGBM Course: UROP1100, Spring Indoor localization has always been a significant field for its commercial interests. As GPS signals are feeble inside buildings, other signals such as Wi-Fi are utilized to localize users. Among all the techniques, Wi-Fi fingerprinting has been the most promising one for its high applicability in complex indoor environments. Euclidean distance, cosine similarity, and Tanimoto similarity are widely adopted to compute the user’s location in the deterministic method. Aside from previous mathematical methods, Autoencoder also has good applications in similarity search with its dimension deduction. Hence, the research team explored the possibilities of adopting Autoencoder in the Wi-Fi Fingerprint-Based Indoor Localization field. The report aims at demonstrating the performance for Indoor Localization implemented with Autoencoder. Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: GUO Zhi / SENG Course: UROP1000, Summer During the three months in the summer term, I have been doing the UROP1000 project: Indoor Localization and Mobile Computing by professor CHAN Gary Shueng Han. In this research course, I have learnt a lot of things about doing research. I tried to read and understand papers on different topics for the first time. I found that many techniques used in the papers are totally new to me. So, I worked hard to learn this knowledge then reread those papers, trying to have a better understanding of them. Although I didn’t really accomplish anything useful in this project, the knowledge I got during this process still means a lot to me. Here are the details. Indoor Localization and Mobile Computing Supervisor: CHAN Gary Shueng Han / CSE Student: ZHANG Zhong / SENG Course: UROP1000, Summer This report analyzes and compares different methods that decode a rasterized format floor plan image. The work aims to sampling reference points on floor plans automatically. Methods including simple image processing algorithms and deep learning are discussed. Some of them are implemented and further tested using UST indoor map data. The report analyzed their advantages and disadvantages, discuss their usage, and potentially enlightening in indoor localization. In the end, I raised a method that integrated machine learning and simple image processing to do this work. It may help and bring some idea to the future development of automatically floor plan image processing.