Autonomous Distraction Detection System

Overview

Every 13 seconds, a life-altering car accident occurs. Despite the hope placed on Autonomous driving, it remains in its early stages and currently is not the solution. That's why I developed the Distraction Warning System. This program detects distractions in both autonomous and human-driven cars, significantly reducing accidents and resolving this critical issue.

My solution

As distractions are imperative factors in these life-altering accidents, my solution is to implement a computer vision model that uses machine learning to identify a phone and track the motion of the hand. Thereby using each of these positions to see if the hand is near to the phone, distracted or not. 

Software Setup:

I used python as it is my natural language. Python 3.8 was used with Anaconda environment.

More description about the software is at the end with a GitHub link to the repository.

Hardware Setup:

Camera: Microsoft LifeCam HD 3000

Hardware: My computer, Lenovo Idea pad 3, with low performance CPU & GPU

Experimental model (actual solution would not have the same hardware or setup):

Result & Conclusion

After multiple data recordings and analysis of both the raw and tampered data, there are a few cases were it misidentifies the distraction especially when a user/driver rotates the steering wheel. From my data sets analysis, this was caused due to the driver's hand closely near to the phone, which causes false-alarms. However, after the experiments and data analysis it, the program did successfully identifies when the driver is distracted, particularly when the driver is texting and driving. Improvements can definitely be made to update and better the performance of the program including the software, and most importantly the hardware, as I was not able to have great expensive hardware to perform my software.

To view the code