COINnoisseur

What it does

Our mission is to simplify and shorten the process of finding valuable coins for collectors and cash-heavy businesses through the efficient detection of these coins.

Your inspiration

Our sponsor's father was an avid coin collector, so he came to us with a vague idea of a coin valuation device in honor of his father. We were touched by his passion and story, so we decided to take up the project and design a novel way to identify high-value coins.

How it works

The COINnoisseur device automates coin valuation using a Raspberry Pi, camera, and custom computer vision pipeline. Coins travel down a ramp and are photographed in real time. Images are preprocessed to extract edges and features, then analyzed by a neural network trained on over 14,000 coin images. The system detects the coin type, corrects rotation, and classifies the date using OCR techniques. It flags high-value coins based on a database of collectible mint years. With a >93% accuracy rate and speed of over 80 coins per minute, the device is compact, reliable, and affordable. This is ideal for collectors and cash-heavy businesses.

Design process

We began by interviewing our sponsor to define the vision: a device to identify valuable coins efficiently. Our initial prototype used a webcam and a basic Python script, but it lacked speed and precision. We then designed a mechanical system with a funnel, ramp, rails, and chain drive to guide coins smoothly. A Raspberry Pi and camera captured images, while a neural network trained on 14,000+ images handled coin type, rotation, and date recognition. We iterated on lighting, coin alignment, and software accuracy, reaching 93% overall precision and processing 80+ coins per minute. Improvements included UI enhancements and better hardware stability. The final device is compact, reliable, and cost-effective, built for under $1000.

How it is different

Devices currently exist on the market that can sort large volumes of coins, but they are unable to determine if they are rare (worth more than their face value). Coin appraisal apps fall short since they have low throughput and cannot scale to the demands of businesses and serious collectors.

Future plans

Our future plans include adding dime compatibility to expand use cases and acquiring more training data to improve model accuracy. We also aim to develop features that detect coin misprints and assess condition/grade, which are critical for collectors. Lastly, we plan to refine the design for easier maintenance, making the device more user-friendly and durable for long-term use.

Awards

Our invention won the "Best Conceptual and Computational Design" Award at the Huff OEDK Engineering Design Showcase at The Ion in Houston, Texas, on April 17, 2025. We were awarded a cash prize of $500 and a certificate.