StreetSignSense is a Machine Learning and Object Detection project focused on real-time identification and classification of traffic signs. The project explores the potential of executing Artificial Intelligence models directly in the browser (client-side) to ensure low latency, privacy, and high performance on edge devices.
The web application is entirely hosted on GitHub Pages and demonstrates the model's capabilities in a real and accessible environment.
The repository covers the entire Machine Learning lifecycle, structured into three critical phases:
Use of Ultralytics YOLO (12) for supervised training on a large-scale traffic sign dataset.
- Engine: PyTorch + Ultralytics
- Dataset: Kaggle (Identified via DOI)
- Output: High-precision
.ptmodels (optimized mAP).
The heart of the innovation: running the model directly in the user's browser.
- Framework: TensorFlow.js with WebGL/WASM backend.
- Logic: Image pre-processing (resize, normalization) and Post-processing (Non-Maximum Suppression) implemented in pure ES6+ JavaScript.
- Frontend: Reactive interface for real-time video stream management (
getUserMedia).
- Zero-Latency Network: Inference happens on-device (
Edge Computing), eliminating network delays. - Privacy-First: No video data ever leaves the user's device.
- Cross-Platform: Compatible with any device equipped with a modern browser (Chrome, Firefox, Safari, Edge).
- Robustness: Trained to handle variations in lighting, angles, and partial sign occlusions.