We developed a CNN based model which classifies a given image into one of the ten classes defined for distraction or safe driving based on the activity the driver is doing in an image. We used StateFarm dataset available on Kaggle for our project. For getting real time inference from the trained model, we used Apache Airflow to orchestrate our Machine learning pipeline.
An example image of a distracted driver
Project Proposal can be found here
Project Report can be found here
Evaluation using Tensorboard
Confusion matrix
Team Contributions
- Arpitha Gurumurthy - Model Deployment on Airflow, Visualization of results, Documentation
- Surabhi Govil - Preprocessing images, Model Architecture, training CNN and VGG models, Documentation
- Gayathri Pulagam - Preprocessing images, Model Architecture, training CNN and EfficientNet models, Documentation
Reference for docker installation
Steps for docker
- bash <(curl -s https://get.docker.com/)
- sudo docker build -t driver-drowsiness:latest .
- docker run -d -p 8080:8080 -p 8008:8008 driver-drowsiness
It'll bring up the docker. - bash scripts/nginx-airflow.sh
On browsing to the url - 35.193.94.65, airflow UI shows up. - bash scripts/nginx-app.sh
On running the above command, our application UI shows up.
Reference for python and UI - https://github.com/krishnaik06/Malaria-Detection