Advanced Machine Learning & Computer Vision
Real-time computer vision systems featuring advanced hand tracking, gesture recognition, and pose estimation algorithms. Implementing cutting-edge machine learning techniques with OpenCV and MediaPipe for high-performance human-computer interaction.
Machine learning-powered sign language recognition system capable of detecting and classifying American Sign Language (ASL) alphabet gestures in real-time. Built using MediaPipe hand tracking, OpenCV, and Random Forest classification for accurate gesture recognition.
The system uses MediaPipe's 21-point hand landmark detection to extract normalized hand position features, which are then classified using a trained Random Forest model. The pipeline includes data collection, feature extraction with landmark normalization, model training with scikit-learn, and real-time inference with live camera feed processing.
collect_imgs.py - Automated dataset generation capturing 100 images per class across 26 ASL letters using OpenCV camera interface
create_dataset.py - MediaPipe hand landmark detection with normalized coordinate extraction for position-invariant recognition
train_classifier.py - Random Forest classifier (200 estimators) achieving high accuracy on ASL alphabet recognition with train/test validation
inference_classifier.py - Live camera feed processing with hand landmark visualization and letter prediction overlay at 30+ FPS
AI-powered web application that uses advanced computer vision to aid filmmakers and content creators in analyzing cinematic footage. Features intelligent scene detection, shot composition analysis, and automated visual insights for film production workflows.
The application leverages state-of-the-art computer vision algorithms to provide real-time analysis of film footage, helping cinematographers and directors make informed decisions about shot composition, lighting, and visual storytelling. Built with modern web technologies for seamless accessibility and performance.
Advanced computer vision systems implementing real-time hand tracking, gesture recognition, and pose estimation algorithms. Demonstrates expertise in machine learning, computer vision, and real-time processing optimization using industry-standard frameworks.
Real-time finger joint detection and tracking using MediaPipe framework with 21-point hand landmark recognition achieving sub-pixel accuracy. Custom gesture recognition algorithms enable intuitive human-computer interaction with 30+ FPS performance on standard hardware.
Built with OpenCV, MediaPipe, and NumPy for advanced image processing. Custom gesture classification models with multi-threading optimization for real-time processing. Achieved 95%+ gesture recognition accuracy across multiple hand positions and lighting conditions with <50ms latency response time.
Gesture recognition accuracy across multiple hand positions and lighting conditions
Consistent processing speed with low-latency response (<50ms)
Sub-pixel accurate hand landmark recognition and tracking