Computer Vision Toolkit for Calibration, AR, and Matching
Implemented end-to-end camera calibration using chessboard images, estimating intrinsic and extrinsic parameters and lens distortion, and performing image undistortion with OpenCV. Developed stereo disparity map generation using StereoBM and SIFT-based feature detection and matching, integrated into a PyQt5 GUI.
University Chatbot using RAG (Retrieval-Augmented Generation)
Implemented a RAG-based chatbot for university-related queries by integrating vector embeddings, similarity search, and LLM inference. Indexed university documents using a vector database for context-aware question answering.
AI Adaptive Traffic Signal Optimizer
Built an AI traffic signal optimizer for Indian cities using PPO reinforcement learning and SUMO, coordinating three intersections to reduce average delay by approximately 30% and outperform fixed-time signals in mixed traffic. Created a prototype with GPS heatmap generation, custom Gym environment, TraCI integration, adaptive baseline, and Streamlit dashboard.
Object Detection and Image Classification with Deep Neural Networks
Trained a Faster R-CNN object detector on the Pascal VOC dataset and performed end-to-end inference. Implemented a ResNet-18 image classifier on CIFAR-10, tracking training and validation loss and accuracy and selecting the best checkpoint based on validation performance.