Carmelo Scribano, Giovanni Cappelletti, Elia Giacobazzi, Giorgia Franchini, Paolo Burgio and Marko Bertogna
To be presented at RAGE worksop@CpS Iot Week.
This is the repository that accompanies the paper "Low-Latency Embedded Driver Monitoring System with a Multi-Task Neural Network".
The proposed model takes as input a
- 98 facial landmarks
- Eye Visibility
- Eye Opening level
- Mouth Opening level
- Head Rotation
- Engagement in distracting activi (Smoking, Cellphone Usage).
Please refer to the paper and the inference code for further details.
The simplified implementation released in this repository serves as a minimal example of the inference pipeline, providing a useful starting point for porting and evaluation across various inference platforms (TensorRT, NPU, Edge TPU, etc.). In this example, OnnxRuntime is used for inference. This implementation is not intended as a benchmark for inference performance; the results reported in the paper are obtained using TensorRT on the Jetson Orin Nano platform.
The reference implementation use Ultra-Light-Fast-Generic-Face-Detector-1MB for face detection and an implementation of SORT for tracking.
We are preparing a form to distribute the trained model to anyone who is interested. Weights will be available solely for non-commercial research purposes.
All content in this repository is released for non-commercial use under the CC BY-NC 4.0 license.
If you found this material useful and would like to use it in your work, please consider citing our paper.
The reference will be added after publication