DQube

Inspiration

Accidents due to human error are wholly preventable and measurement of how various distractions can lead to accidents can help in raising public awareness of their actions

What it does

Detect how the driver is engaging in distracted behavior and alert in case of distraction.

How I built it

Implementing VGG_16 neural network architecture on the dataset to predict for ten different classes of distraction. Using OpenCV for live testing from the camera feed.

Challenges I ran into

Building a machine learning model and setting up the tech pipeline for a large number of unstructured images. Feature engineering for the dataset.

Accomplishments that I'm proud of

The model has enough flexibility to detect behavior in images taken from a mobile phone.

What I learned

Efficient feature engineering can help in kick-starting the model building

What's next for DQube

This model can be used as input to measure the distraction class and combining the time of distraction, the vehicle features, and geo features can help in understanding the propensity of an accident for a driver. It can raise public awareness for the prevention of accidents and can also be integrated into dashcams to automatically alert drivers.

Built With

• keras
• opencv
• python
• scikit-learn