We collaborated with BDTI and TryoLabs to create MaskCam, a real-world demo showcasing the NVIDIA Jetson Nano’s deep learning capabilities.
A mask detection camera prototype based around the NVIDIA Jetson Nano.
EJ Technology Consultants was contracted by BDTI, a technology development and evaluation firm that specializes in embedded computer vision applications, to assist with evaluating the NVIDIA Jetson Nano’s suitability to real-world edge AI applications. The demo was a mask detection smart camera that we developed from an initial concept into a production-ready prototype. A video showing MaskCam in action is available on our YouTube channel.
MaskCam detecting masked faces at the airport.
MaskCam utilizes NVIDIA’s TensorRT library to implement a high-speed vision pipeline that runs live camera frames through a YOLOv4-tiny model trained to detect masked and unmasked faces. The optimized pipeline runs at 14FPS on the Jetson Nano. The program reports mask wearing statistics to a web server and saves recorded video if the number of unmasked faces is above a certain threshold.
The deep learning image processing pipeline running in MaskCam.
EJ Technology Consultants contributed to the system design of the smart camera:
- Wrote top-level system requirements and product specifications
- Evaluated and selected hardware that met specifications while maintaining a low cost target
- Integrated hardware drivers with embedded Linux operating system and troubleshot software/firmware where necessary
- Performed system testing
A browser-based interface allows users to view detection statistics from mulitple camera devices.
We collaborated with a cross-functional team of software, DevOps, and production engineers from TryoLabs, BDTI, and Jabil during the development process. We also presented with BDTI and Tryolabs at the NVIDIA GTC and Embedded Vision Summit conferences.