Real-Time Object Detection is a high-performance computer vision pipeline written entirely in Python. It captures live webcam video using OpenCV, runs image pre-processing instantly via NumPy, and passes the data into an optimized, local ONNX Runtime engine containing a quantized YOLO model.

Source code, dataset, and hyperparameters for the Melipona capixaba detection and tracking (MOT) pipeline, evaluating trade-offs between GPU processing and Edge Computing. (WCAMA 2026)

Active perception system for target localization on a mobile robot— maintains a Bayesian belief map over object location, plans viewpoints by mutual information maximization, and executes via behaviour tree in ROS 2.

browser-based video inspection and benchmarking tool for object-detection models. It lets users run YOLO/ONNX inference directly in the browser, inspect detections frame by frame, compare model confidence over time, manage labeled scene clips, and review run history locally without a backend.

Custom-engineered computer vision solution that goes beyond simple detection to provide a deep understanding of scene physics