@vechile login system using web and normal cam

Browser-Based Browser-Based Vehicle Attendance System GateWatch is a vehicle gate management system that identifies vehicles by their number plate and logs check-ins and check-outs against a registry — using nothing but a standard camera and a web browser. There's no dedicated ANPR hardware, no app to install, and no cloud vision API in the loop; the entire recognition pipeline runs client-side. The core challenge wasn't reading text off an image — it was building a recognition pipeline reliable enough to act on automatically, using footage that's rarely clean. Real gate cameras capture plates at odd angles, in poor light, partially obscured, and in motion. GateWatch addresses this in layers. A custom computer-vision routine scans the live video feed for plate-shaped rectangles in real time, using edge detection and stroke-density scoring rather than a heavyweight ML model. Once a plate is located, the crop is denoised, sharpened, and upscaled before being passed to a persistent OCR engine (Tesseract's LSTM recognizer, kept warm for speed). Rather than trusting a single OCR pass, the system reads each plate multiple times across different segmentation strategies and takes a confidence-weighted vote, discarding low-confidence attempts outright. A final correction step fixes common character misreads (like O/0 or S/5) — but only at positions where the plate's own format structurally guarantees the correct character, so nothing is ever guessed. The result is a lightweight, self-contained system: point a camera at a vehicle, and GateWatch locks onto the plate, reads it, cross-checks it against a registered vehicle list, and records the entry or exit — all in the browser, in real time. Beyond the technical build, GateWatch is a small case study in a broader principle: reliable automation isn't just about getting a model to produce an answer, but about building the judgment to know when that answer shouldn't be trusted. Beyond the technical build, GateWatch is a small case study in a broader principle: reliable automation isn't just about getting a model to produce an answer, but about building the judgment to know when that answer shouldn't be trusted