People Detection, Tracking and Facial Blur
A pipeline using Nvidia PeopleNet to detect persons and faces. Faces are matched to bodies and blurred with OpenCV CUDA. Includes a tracker to reduce box flickering.
Key Features
- Body and facial detection with PeopleNet
- Facial blur with OpenCV CUDA
- People tracking with Nvidia Tracker
- Embedded analytical dashboard
Performance
- 116 FPS on Nvidia A4000
- 30 FPS on Jetson NX
This introductory sample demonstrates Savant’s core capabilities for person detection, tracking, and privacy-focused facial blurring using NVIDIA’s PeopleNet model.
The pipeline detects both bodies and faces, matches faces to their corresponding bodies, and applies GPU-accelerated blurring through OpenCV CUDA integration. A simple tracker reduces detection box flickering across frames.
Getting Started
git clone https://github.com/insight-platform/Savant.git
cd Savant && git lfs pull
docker compose -f samples/peoplenet_detector/docker-compose.x86.yml up
Open rtsp://127.0.0.1:554/stream/city-traffic in a player or visit http://127.0.0.1:888/stream/city-traffic/ for LL-HLS.