Traffi is a proof-of-concept computer vision project to analyze traffic intersection crossings.
At this intersection it detects illegal right turns made by westbound traffic, and illegal left turns made by eastbound traffic on Grand St.
We’re also building a dataset of low resolution bird’s eye views of traffic intersections, to train vision models for efficient, privacy preserving monitoring of community traffic.
The live feed is generated and served from a Raspberry Pi 5 equipped with an inexpensive camera and a HAILO 8L AI accelerator. It detects vehicles using an artifical neural network, and combines a bit of simple math to deduce which side they originate from, and eventually, migrate to. It only consumes 7W.
Developing the dataset and training the AI model to increase detection accuracy is our current goal. We’re also trying to secure funding to support community monitoring and dataset collection at more intersections.
NOTE: the counter shown in the top left of the video is a rudimentary early feature, and may reset from time to time. Please refer to the Stats page for the most accurate statistics available.
This idea started as a learning project in leveraging edge AI computing hardware. The safety of this traffic particular traffic intersection has been a key issue in the local community.
As such, it’s hoped this project can also serve the community by informing and arming them with real data on the issue.
Please use the feedback form to share your questions, feedback or to express interest in working with us.