Inspiration

Any country finds a core of its GDP in the Industry sector, which contributes 24% to our country's GDP while employing millions of people in blue-collar roles in medium and large industries. What unites majority of roles across the Industry sector are the long hours of commitment and tasks like injection moulding or long distance truck driving - activities that also require high amounts of attention and focus because they are risky, with an active threat to the employee's life in case of a mishap.

We want to remedy the likelihood of this ever happening; we want to help the employees stay alert and focused at work while ensuring that the employers are aware of how their employees are doing and holding up at work through an effective fatigue management system, thereby providing for a safer work environment which will see accidents as a regular occurrence.

What it does

The application uses Machine Learning (trained on AWS) and Computer Vision, to track the drivers face movements and cross validates the probability of the driver being fatigued. The factors to detect a drivers fatigue are- eye motion, head orientation, heart rate and time since last pit-stop. If the driver is found to be fatigued, the team crew are informed about the information, such that the car can visit the pit-stop and driver can either be replaced or provided support for optimal performance. The race crew also receives Video Footage of the drivers fatigue for physiological analysis after the race.

How we built it

Analyzing driver video feed using OpenCV and providing Computer Vision for tracking and point mapping. The data is fed into a GridSearch Model with kernel as 'rbf'. The model was trained on AWS using self recorded data set. The accuracy of the system was found to be 87.2%. The advantages of using such a system is that the machine can be fed actual data about driver fatigue such that it can be retrained for greater accuracy.

Challenges we ran into

Identifying the existing problem in the racing industry. Initializing OpenCV to identify video footage and mapping points using dlib. Machine Learning training was another challenge due to lack of data set, so we had to generate points using manual instances.

Accomplishments that we're proud of

Producing a machine learning model with a dataset that can identify driver fatigue with such great accuracy.

What's next for the Project

Implementation for Consumer and Commercial scales. Porting the entire platform on cloud and mobile devices for greater use case. Providing cloud runtime for off device compute.

Built With

Share this project:
×

Updates