Brand-Detect

Inspiration

I wanted to created a way to track the popularity of car brands on road from traffic data

What it does

Uses Yolo image detection to detect cars on the road - then uses a crop so the LLM (Claude Sonnet) can make an educated guess on what brand the car is or even what model of car.

How we built it

I used Sagemaker to host my AI module - I used a S3 bucket to pool data from the Inrix data set of traffic camereas. I then also used AWS lambda function to give user input so you can input your own traffic camerea images for scanning.

Challenges we ran into

Low quality camerea images - there isnt much of a work around for that.

Accomplishments that we're proud of

The Sonnet Module is actually very aware of the differnces between car types - I was very impressed with its judging capabilities. So I was happy that feature seemed to pull through.

What we learned

I need to be more aware in the future of the limits of my AWS permissions I lost a lot of time becuase of this.

What's next for Brand-Detect

I want to train my own vision brand detection module instead of relying on a LLM.

Built With

• amazon-web-services