Taking attendance during university exams is a universally inefficient process. Invigilators waste the first 15 minutes of a critical testing period walking around with clipboards, checking IDs, and matching faces to names. It is a manual data-entry problem masquerading as a security protocol. I wanted to automate this completely. The goal was to build a system that could passively scan students as they entered the hall and log their attendance in real-time, requiring zero manual input. The result is Project Nero—a local facial-recognition attendance tracking system. Here is how I architected the computer vision pipeline to make it fast, accurate, and completely independent of expensive cloud APIs. The Tech Stack Streaming live video to a cloud AI endpoint is slow, bandwidth-heavy, and raises massive privacy concerns for student data. To solve this, Nero processes everything on the edge (locally) using: Core: Python Video Processing: OpenCV Facial Recognition: DeepFace / MediaPipe Data Mapping: External Spreadsheet Integration The Architecture Pipeline 1. Frame Capture & Face Detection (OpenCV) The system initializes a local Python camera script that hooks directly into the hardware webcam. Using OpenCV, the script captures the live video feed and breaks it down into individual frames. Before we can identify who is in the frame, the system must figure out where the faces are. I utilize a lightweight cascade classifier to isolate bounding boxes around human faces in real-time, cropping the frame down to just the necessary pixels to save processing power. 2. Facial Categorization (DeepFace) Once a face is isolated, the cropped image is passed to the DeepFace library. DeepFace acts as the core recognition engine. Instead of retraining a model from scratch, DeepFace wraps state-of-the-art models to generate a high-dimensional facial embedding (a mathematical representation of the face's features). It compares this real-time embedding against a pre-registered local directory of student photos. If the similarity score breaches the confidence threshold, the system successfully categorizes the face and returns the student's unique ID. 3. Real-Time State Mapping (Spreadsheet Integration) Recognizing a face is only half the battle; the system needs to log the data. Once DeepFace returns a matched student ID, the Python script triggers an external spreadsheet integration. It maps the recognized ID against the master student roster, flips their attendance state from "Absent" to "Present," and logs the exact timestamp of entry. To prevent the system from spamming the spreadsheet with 30 requests per second while a student stands in front of the camera, I implemented a local state-cache. Once a student is logged, their ID is cached in memory, and the system ignores them for the rest of the session. The Result Project Nero eliminates the manual bottleneck of exam invigilation. By combining OpenCV for rapid frame processing and DeepFace for accurate categorization, the entire pipeline runs locally on standard hardware. It is privacy-conscious, requires no cloud compute costs, and completely automates the administrative overhead of tracking students. I’m Ahmad Akmal (RuumiDev), a software developer specializing in interface design, systems architecture, and practical AI. If you prefer structured, practical engineering, let's connect: GitHub: github.com/RuumiDev Website: ahmadakmal.dev