Oshan Mudannayake

Selected Projects

  • Water Surface Detection from UAV Imagery

    UNet, TensorFlow, OpenCV, Python
    UNet-based semantic segmentation pipeline for water-surface detection in low-altitude UAV imagery; published in JVCI (Elsevier).
  • Municipal Solid Waste Generation Forecasting

    Time-series ML, Darts, TensorFlow, Python
    Multi-step time-series forecasting of municipal solid waste generation benchmarking classical ML against deep sequence models (LSTM, N-BEATS, TCN); published in IEEE Access.
  • Realtime Property Evaluation of Large Streaming Graphs

    Streaming Graphs, Graph Summarization, Python
    kMatrix — a space-efficient sketch for streaming graph summarization supporting real-time structural property queries over unbounded edge streams; published at IEEE ICIAfS; basis of BSc thesis (Grade A).
  • ImageLab

    Python, OpenCV, Image Processing
    Interactive framework for composing, visualizing, and inspecting image-processing pipelines; co-first-author manuscript under peer review.
  • Machine Learning over Encrypted Data

    ML, Homomorphic Encryption, TensorFlow, Python, GCP
    Privacy-preserving training layer over TensorFlow enabling model fitting on homomorphically encrypted tensors across a distributed worker network; research-intern project at UCSC DCG.
  • TessTA

    Tesseract, OCR, React, Java, Spring Boot, AWS
    OCR system for low-resource Sinhala script with a custom-trained Tesseract model and a browser-based annotation/correction interface for iterative dataset curation.
  • Intelligent Traffic Light System

    TensorFlow, CNN, Raspberry Pi, Arduino, MQTT, Angular, NodeJS, GCP
    Replaced fixed round-robin traffic-light scheduling with a vision-based adaptive controller using on-device CNN vehicle detection on Raspberry Pi, coordinated across intersections over MQTT.
  • Smart Glove

    FSR-400, IMU, Arduino, Java, JavaFX
    Instrumented glove prototype (FSR-400 pressure array + IMU on Arduino) capturing hand pressure distribution and acceleration to characterize cricket batting grip dynamics.
  • Yarn Detection System

    OpenCV, Python, Raspberry Pi
    Embedded computer-vision system on Raspberry Pi detecting knots in yarn threads on industrial glove-weaving machines, eliminating hundreds of broken needles per day and reducing manual inspection.
  • Glove Defect Identification System

    OpenCV, Python, Image Processing
    Conveyor-line defect detection for textile gloves using classical CV, enabling automated rejection of defective units without operator intervention.
  • Bra Elasticity Analyzing System

    OpenCV, Java, Swing, Image Processing
    Image-processing pipeline quantifying and comparing elasticity of sports bras from deformation imagery under controlled loading.
  • Machine Learning based Internet Domain Entity Matching and its Applications

    ML | Scikit-learn | Tensorflow | Python
    Aimed to predict the malicious domain URLs by using machine learning techniques.