DistractedDriverDetector ~ A Neural Network Approach

• 

  Overall workflow flow chart.

• 

  Neural Network Flow Chart (hidden layers are 128 neurons).

• 

  Output flow chart.

Inspiration

According to the National Highway Traffic Safety Administration, 3,142 people died in distracted driving accidents in the US in 2022. Additionally, distracted driving increased by 20% from 2020 to 2022. Stop lights have cameras that detect whether a car has run a red light, and speed limits have speed detectors that can track if a car is speeding, but what system is currently in place to monitor distracted drivers? Phones have the technology to monitor if a driver is using it while driving, but what about when a driver has their eyes off the road and their attention is not on driving? Our team set out to resolve this issue by creating a novel application that can determine if a driver is distracted through photos of them while they are behind the wheel.

What it does

This application identifies distracted drivers in real-time. It can take images captured by the camera and/or uploaded from the gallery. With these images, the application outputs whether it is an instance of safe driving or distracted driving. If it is distracted driving we ensure human verification (by the user) such that a false report is not made. It can be used in the real world by police officers to actively identify distracted drivers and create and/or store additional information in their permanent records (explained in the output workflow figure). And in the case, the driver wants to argue the fact that they were distracted, the police officers will now have concrete photographic evidence of their case.

How we built it

The general workflow is displayed in the general workflow figure. The application was built through Google Colab for the Neural Network (explained in the neural network figure) and Android Studio for the Android application. Through Google Colab we imported a data set consisting of distracted drives and safe drives. The written neural network was then trained on this data and saved to be imported into Android Studio. On Android Studio we allow the user to choose between taking a picture or uploading a picture file. This image is then sent to the trained Neural Network which outputs whether it is safe driving or distracted driving and the process is continued (explained in the output workflow figure).

Challenges we ran into

We ran into challenges in optimizing the Neural Network to recognize distracted driving while keeping the training time reasonable. We also had challenges sharing information between the Neural Network and the Android application. We tackled these challenges by developing a deeper understanding of convolutional and dense neural networks, and a better understanding of TensorFlow’s integration with Android development.

Accomplishments that we're proud of

We are proud of the interface we created through Android Studio with camera access so users can take pictures themselves, and photos access so that users can upload photos they have. We are also proud of the accuracy of our neural network as it boasts a 99.3% accuracy on the trained dataset. We are also proud of integrating human verification to ensure we are not solely relying on artificial intelligence but simply making it easy for law enforcement to use the application. Above all, we are proud that we have developed an application that can be used in the real world to promote safer driving practices and therefore save countless lives.

What we learned

Throughout our experience developing this application, we learned many skills including but not limited to Android App Development, Neural Network Creation, and Optimization. As we progressed in our project, we developed the skills to solve real-world problems through practical and implementable solutions. In addition to these, we all also improved upon soft skills like communication, team building, and teamwork.

What's next for DistractedDriverDetector

The next step for the DistractedDriverDetector is getting it integrated with law enforcement, to send citations or warnings to inevitably reduce the amount of distracted driving, and distracted driving accidents. DistractedDriverDetector’s next step is making the world a safer place. DistractedDriverDetector when integrated with other advanced traffic management systems can also help in traffic optimization, and effective road utilization and will help local, state, and federal governments in need.

Built With

• java
• python