Designing Intelligent Surveillance with YOLOv11
Building a high-accuracy person detection system for video surveillance using YOLOv11, GPU acceleration, and custom training pipelines.
The problem
Video surveillance pipelines often trade off accuracy for speed, missing people in cluttered or low-light footage, or running too slowly to keep up with live camera feeds. The goal was a person-detection system that stayed accurate under real-world conditions without needing specialized hardware.
What I built
- Built a high-accuracy person detection pipeline on YOLOv11, with a custom training pipeline tuned for surveillance-style footage (varied lighting, occlusion, camera angles).
- Added GPU acceleration to keep inference fast enough for live or near-real-time camera feeds.
