Offline Car-Plate & Sign Detection

I need a complete, offline-ready computer-vision pipeline that can reliably detect vehicle licence plates and traffic signs in real-time video or still images. Everything must run locally without any cloud calls, built on TensorFlow and Keras with only open-source components. My goal is to receive a solution that I can drop onto an edge device or standard PC, start a camera feed, and immediately see bounding boxes around plates and recognised traffic signs. Model accuracy must be on par with commonly cited public weights; speed should be good enough for at least 15 fps on a mid-range GPU (CPU-only fallback appreciated). Deliverables • Fully documented source code in Python using TensorFlow + Keras • Pre-trained weights (or clear training script plus instructions to reproduce) for licence-plate and traffic-sign classes • Simple CLI or lightweight GUI to load a video / camera stream and export detections (JSON or CSV + annotated frames) • Setup guide that lets me install and run everything offline on Windows or Linux in a single sitting Acceptance criteria 1. Detection precision/recall demonstrated on a short sample set I provide 2. Runs entirely offline; no external API calls, licence servers, or closed-source binaries 3. Clear, step-by-step documentation given; I can reproduce your test results on my hardware within one hour I’m ready to test as soon as you share an initial build, and I’d like the final package ASAP, so please let me know your expected turnaround and any dependencies you foresee right away.

Project ID: 40269252