SecureCam

Conventional webcams capture video but typically lack intelligent monitoring, automated event reporting, and convenient remote management. Continuous recording can generate unnecessary footage, while users may need timely notifications when someone enters a monitored area.

The objective was to develop a security monitoring application that detects people using computer vision, records meaningful detection events, and provides centralized access to camera controls, snapshots, and alerts through a web dashboard and Telegram.

Implemented a Python camera agent using OpenCV for video capture and a pretrained YOLOv8n model for person detection. Added configurable confidence thresholds and automatic CPU fallback when CUDA acceleration is unavailable.

Designed a presence-state event handler that triggers an event when the scene changes from empty to occupied. Introduced an absence grace period and configurable cooldown to prevent repeated alerts during a single visit. Added local snapshots, optional video clips, and asynchronous event reporting to avoid blocking the camera loop.

Developed a REST API using FastAPI and SQLAlchemy to manage users, paired devices, detection events, and snapshots. Implemented JWT authentication, separate device tokens, expiring single-use pairing codes, and ownership checks for protected resources.

Used WebSockets to synchronize device state and deliver remote configuration changes. Built a Next.js and TypeScript dashboard for managing devices, reviewing event history, viewing snapshots, and controlling camera monitoring.

Integrated Telegram's Bot API with OTP-based account linking, detection notifications, device listings, and remote arm/disarm commands. Added configuration for PostgreSQL-compatible deployment and an automated Windows agent build workflow.

Developed an integrated AI security monitoring application consisting of a Python detection agent, a FastAPI backend, and a Next.js web dashboard.

Implemented person detection with YOLOv8n, presence-based event triggering, snapshot capture, optional clip recording, device pairing, remote camera control, and Telegram notifications.

Built real-time communication between the backend and camera agents, allowing users to manage monitoring status and supported device settings remotely.

Added authentication, device-specific credentials, event history, and ownership-checked snapshot access. The repository also includes deployment configuration and a GitHub Actions workflow for packaging the Windows camera agent.

The source code demonstrates an end-to-end monitoring architecture; production reliability, detection accuracy, and inference performance have not been independently benchmarked.