Inspiration
Transportation networks are at the heart of all cities, they enable the movement of goods and people across the city through various means. For road users which include pedestrians and drivers, a traffic light will be inevitable in their journey. The average American will spend about 4.7 months over the course of their lifetime waiting at a red traffic light. link. Everyone has found themselves at least waiting at a red light unnecessarily when there is no traffic.
The traditional traffic light design involves data analytics to predict road demand to determine signal timing. This has two problems: first, you need to collect data. Second, it is a passive system that does not react to its surroundings.
Current design solutions in transportation involve sensors of various kinds to anticipate incoming vehicle traffic. They mainly give priority to public transportation units such as buses to increase signal timing.
The project name comes from another initiative named vision zero, which aims to reduce traffic collisions to a total count of zero. This initiative aims to reduce unnecessary traffic congestion to a count of zero as well.
What it does
We've created a reactive traffic signal light intersection using computer vision and machine learning to create a feedback loop to traffic lights. This vision of incoming traffic will inform the traffic light of a decision to make upon recognizing its current surroundings.
This solution benefits all road users, including pedestrians. As the traffic lights can now react to pedestrian flow as well. In addition, it's noted that with this software implementation, data can now be collected if needed to do further analysis and transportation studies.
The flexibility of this solution enables transit planners to still develop a transit algorithm that they deem fit to determine traffic light signals while enabling it to react to the surroundings. Including but not limited to environmental factors as well.
How we built it
This was built with pyqt5 and python for the GUI interface. Darknet is the main machine learning model algorithm for visual object detection. Tiny -YOLOV2 was chosen to help with the speed and processing due to CPU limitations in processing images. OpenCV is used to overlay the boxes of object detection overlayed on top of the images. Imutibl python package was used to process the videos into frames for analysis.
Challenges we ran into
Due to the limitation of GPUs, the processing speed is a bit slow. Also, the configuration of darknet proved more difficult due to outdated packages and being innate to Linux machines as opposed to windows machines.
Accomplishments that we're proud of
A fully functional hack that will impact transportation users. This will help reduce carbon emissions as well caused by idle waiting. Have the potential to prioritize pedestrians. And shape a better world in enabling better transportation networks.
What we learned
Computer vision development, and how machine learning plays into this. The future of transportation. For two of us, this is our first hackathon so definitely a learning experience!
What's next for Vision0
To enable faster speed processing, the next development would be to bring it to a web app development and connect to cloud processing for faster processing.
Log in or sign up for Devpost to join the conversation.