Role

Sole Developer: full ownership of design, implementation, and optimization

Course

CPS4893 – Machine Learning & Computer Vision

Objective

The goal of this project was to build an app capable of detecting and correcting human sitting posture using real-time camera input. The app was expected to meet commercial-grade frame rate standards, provide a clean user interface, and maintain low memory and latency consumption during posture classification.

Technologies

Real-time video processing with OpenCV, Human posture classification, MVC pattern design, GPU acceleration with CUDA, UI rendering and frame rate management

My Contributions

• Designed and implemented the full system architecture, UI, and functionality
• Captured and analyzed webcam video in real-time using OpenCV
• Built a posture classification algorithm to detect and correct improper sitting
• Applied CUDA GPU acceleration to optimize prediction speed and ensure consistent frame rate
• Managed system latency and memory usage to meet commercial performance standards

Results

• Delivered a real-time posture correction app with high accuracy and fast response
• Clean UI and immediate feedback enhanced user interaction
• Achieved latency <100ms, suitable for near-commercial deployment
• Gained hands-on experience in building and optimizing computer vision applications

GitHub Repository

Not yet available