SignPatrol – Daniel Greipel

Thank you Quirin Sailer for the great idea and execution allowing me to get hands on experience in creating a real physical product solving a real problem! feel free to have a look at his other successes! (Click on his name)

Study: International Management

I recieved an E-mail from the Center of Entrepreneurship of my University about this Hackadon – An amazing chance to network to developers and deepen my understanding of software and hardware development.

From AI implementation to 3D – Printing and Pitching, everything was part of my exiting learning journey, in which an idea formed to a winning product within three days.

We were able to secure a project related prize money of 3.000€, have the support of our University , the Center of Entrepreneurship (CoE) and of the Challenge initiator Zero GmbH, A local software development company.

My Hackadon Learnings:

1. Concept and Creation of Business Models

2. AI Implementation and Training with real Data

3. Preperation of Engaging pitches for a Jury to understand the value

1. Business Model

The concept was clear quickly – We had the problem at hand: Communes could not know what traffic signs are too dirty to read, but they had the job of maintaining them for traffic security reasons.

So the idea was to create a prototype that is able to learn whether or not a sign was dirty. Using an AI – Cluster approach, we developed a prototype that had a camera and a rasperry pi processor.

I mapped out needs of the communes, and the business model of selling to them. We would have the little hardware boxes that were able to scan the signs on top of public vehicles or taxis and sell a SAAS monitoring software to the communes in which they could see the pictures of dirty signs identified and where on the map these would be.

2. AI Implementation

To train and test our programm, two members went on a drive with one person navigating the way and letting the prototype scan various traffic signs to train the AI into understanding which ones are dirty and which ones were clean.

After a few hours of training, the system was running with over 90% accuracy and we were able to classify the signs into red, yellow and green on a virtual map.

Red: Immediate action, sign not recognizable
Yellow: Sign not clear, often due to stickers
Green: Sign ist clear, no action necessary

3. Pitch Preperation

To prepare for the pitch, we used a sign that the prototype would scan and mark as green when it was clean. Then we would put some stickers on it and let it scan again.

The prototype would send the status of the sign to the interface we were streaming on during the pitch and a sidebar showed the picture of the sign, and yellow, indicating the vandalism.

Finally, we would smear black paint onto the sign, thus making it unreadable and a danger to traffic. The system correctly flagged it as red, securing us first place and the prize money of the hackathon!

My Conclusion

SignPatrol showed how a weekend can become a real product: a small camera + Raspberry Pi that rides along, spots dirty or vandalized traffic signs, and marks them red/yellow/green on a live map. In three days we went from problem to prototype—and earned support from the CoE and Zero GmbH. The takeaway: start with real pain, learn fast from data, and keep the model simple

Help Yourself and me by getting Notion!

This affiliate Link doesn’t cost you more, but I would appreciate purchases using the link as I get commission

Thanks for Reading – Head back to the Homepage to discover more interesting posts!