Role

Involved in all three phases: Group 1: Co-designed project proposal and authored key parts of requirement document. Group 2: Lead developer for data preprocessing, modeling, and evaluation (Crop analysis). Group 3: Responsible for computer vision model optimization and code integration (Pen grip detection).

Course

CPS4951 – Senior Capstone & CPS4981 – Special Topics in Computer Science

Objective

This project comprised three stages: Part 1: Designed a handwriting posture detection system using MediaPipe. Part 2: Built a crop recommendation model using Kaggle dataset and random forest classifier. Part 3: Optimized and completed the pen grip detection system with real-time performance improvements and integrated codebase.

Technologies

MediaPipe, OpenCV, Python, Pandas, scikit-learn

Random Forest Classifier, One-Hot Encoding, Confusion Matrix, ROC Curve, Multithreading, Real-time Video Frame Optimization, Feature Engineering, SHAP Analysis

My Contributions

• Designed requirement document interfaces and coordinated initial planning (Part 1)
• Built preprocessing pipeline: One-Hot encoding, SHAP visualization, feature engineering (Part 2)
• Implemented and evaluated Random Forest classifiers for crop and gesture classification
• Resolved FPS drop and prediction accuracy issues in hand gesture detection
• Integrated the system using multithreading, continuous frame analysis, and unified main function
• Supported teammates in debugging and preparing for final presentation

Results

• Delivered real-time gesture detection with 93% accuracy and smooth frame rate
• Developed crop recommendation engine integrated with UI, validated via confusion matrix
• Successfully demonstrated the full system in poster session and received strong feedback
• Gained comprehensive experience in interdisciplinary collaboration and full-cycle project execution

Project Files

View Full Project Report (PDF)