Undergraduate Projects

Projects Overview

• Enhancing Visual Performance on Dancing Robots with OAK-D PRO
• Exploration of Urban Blight and Property Values via Deep Learning
• Stock Price Forecasting via Machine Learning
• Cleaning Robot Management System
• Image and Speech Processing using Chatgpt-api
• Cyber Security Analysis on APT threat
• Investigating Predator-Prey Model
• Traffic Intersection Simulation
• Snake Game Design with Self-Contructed CPU and Assembler
• Target Shooting Strategy in Stock Market

Enhancing Visual Performance on Dancing Robots with OAK-D PRO

• Conducted research on OAK-D PRO depth camera to enhance the vision performance on dancing robot
• Contructed pipeline using DepthAI’s API in OAK-D PRO camera
• Used openCV to analyze frames detected in the OAK camera and achieve feature tracking and stereo depth estimation
• Utilized Flood IR LED illumination to enable perception in low-light and no-light environments, and implemented Laser dot projector to enhance passive depth perception for low-visual-interest surfaces.
• Utilized openVINO to embed custom model and YOLO model into OAK camera by converting from pytorch form into blob form
• Implemented non-maximum suppression and skeleton plotting
• Embedding YOLOv7 and YOLOv8 model into OAK camera to achieve lower latency object detection and human pose estimation
• Link to the Demo: https://youtube.com/shorts/J5G7IG2wvps

Exploration of Urban Blight and Property Values via Deep Learning

• Conducted research on the influence of house appearance on housing prices using Google Street View photos and pixel-based scoring.
• Employed advanced techniques, including Transfer Learning with VGG16 and AlexNet, for predictive classification of satellite images.
• Utilized R studio's Spatial Data capabilities to establish spatial links between residents and their respective neighborhoods.
• Applied hedonic model incorporating property attributes along with seasonality fixed effects to accurately estimate property value.

Integrated Opinion Leader Sentiment Forecasting for Top Technology Company Stock Prices

• Extracted stock market price data for renowned technology companies fromYahoo Finance.
• Leveraged Twitter's API to extract expert analysis comments on individual stocks, processed irrelevant information, and extracted factors.
• Conducted regression analysis using Linear Regression, Gamma Regression, Neural Network, and LSTM.
• Performed movement classification analysis using Logistic Regression, Support Vector Machine, and K-Nearest Neighbors.
• Demonstrate performance of each model by Utilizing Matplotlib and performance matrix

Cleaning Robot Management System

• MULTI-THREADED TASK MANAGEMENT: We implemented task assignment with multi- threading in C++, enabling scheduling of cleaning operations, charging, and maintenance tasks for robots.
• CLI DEVELOPMENT:We created a C++ CLI for efficient user interaction and operational control of a large-scale cleaning robot fleet.
• DEVELOPMENT TOOLS PROFICIENCY: We showcased expertise in Git and CMake for code version control and build management, highlighting a commitment to software engineering best practices.

Image and Speech Processing using Chatgpt-api

• Utilized ChatGPT-API and DALL-E3 to analyze the input image and output it with different style.
• Provided audio description to the output image with different voices by using Google-API.

Cyber Security Analysis on APT threat

• Conducted research in artificial intelligence and cyber security, topics including intrusion detection and deep neural network architectures.
• Proposed an intelligent threat detection method based on the extended Cyber Attack Chain model and the Long Short-Term Memory network autoencoder to extensively correlate malicious behaviors from spatial and temporal dimensions.
• Publication: A spatial-temporal correlation-based method for advanced persistent threat detection, 2021 4th International Conference on Mechatronics and Computer Technology Engineering (MCTE), Nov 2021
• Link to the publication: https://iopscience.iop.org/article/10.1088/1742-6596/2113/1/012037/meta

Investigating Predator-Prey Model

• Implemented logistics model of predator-prey dynamics using Runge-Kutta methods.
• Conducted parameters analysis using bifurcations and phase plane to identify dynamic relationships between predator and prey.

Traffic Intersection Simulation

• C++ Traffic Intersection Simulation Developed a modular C++ simulation for traffic intersections, emphasizing precision through object-oriented design.
• Code Quality Focus Ensured high code quality with thorough testing, and utilization of advanced features like copy/move constructors.
• Realistic Behavior Modeling Implemented accurate vehicle movement and intersection behavior, incorporating random number generation for realistic vehicle arrival and turning probabilities.
• Link to the Demo: https://youtu.be/y5cMC-loJAQ

Snake Game Design with Self-Contructed CPU and Assembler

• Build a CPU using Logisim, and build an assembler using C++
• Design and implement Snake Game using assembly

Target Shooting Strategy in Stock Market

• Researched and analyzed limit-up/limit-down behavior's impact on stock market returns, employing quantitative analysis techniques.
• Conducted statistical analysis and generated visualizations for cumulative returns across various time frames and five limit patterns.
• Calculated the confidence interval for the factors derived from the analysis and validated accuracy through rigorous backtesting data.
• Developed optimized investment strategy based on key factors, including limit pattern and days after reaching limits.