Project: Machine Vision License Plate Recognition
Client: Fortune 500 Manufacturer
Dunn Solutions is supporting a Fortune 500 manufacturer with a multi-year research and development initiative.
Challenge: Machine Vision Solution Needed for Smart Sign Manufacturing
The challenge for the manufacturer is to enhance their core strength in the material design of signs with computer vision to deliver an integrated solution that will read, understand, and automate workflow around various signs. Signs that could only be identified and processed by humans, such as license plates and pavement markers, can now be encoded and handled using computer software.
Our client was looking for a partner with skills in machine learning and scalable, cloud based, solution and chose Dunn Solutions to partner with in developing and researching new digital products to offer a complete solution to their customers.
Solution: Focus on Smart License Plates
The challenge with license plates is to drive up the accuracy of reading and decoding them as compared to a human or an existing transportation tracking system. License plates typically contain gaps in between the characters and those gaps are nearly invisible to the human eye. Those gaps are however visible to an infra-red camera. Knowing this, our client was able to create specially encoded smart license plates that would be highly detectable through computer vision and would also provide useful data about the vehicle, such as state/province registration, in addition to the plate numbers.
Dunn Solutions architected and delivered a computer vision pipeline using a combination of machine learning algorithms and highly scalable public cloud tools to read the smart license plates, collect detectable data from them and transmit that data to a web based front end user interface. As part of the implementation, Dunn Solutions configured an industrial grade machine vision camera, which utilized an external light source which allows for "floating" plate images to be detected and captured once a vehicle crosses a preset line configured in the camera. Once the vehicle crosses the "invisible line", the camera generates image files which are transmitted to the first stage of the machine learning pipeline. This approach was proven effective over a variety of vehicle speeds and different lighting conditions.
Deep learning modules for plate detection, optical character recognition (OCR) and DSS mark detection were developed and tested as next steps in the pipeline. Technologies leveraged include Amazon Web Services (AWS), Tensor Flow, OpenCV and proprietary algorithms to better automatically review content information on the fly. A website which will showcase the data collected from the smart license plates on a front end user dashboard was also developed.
Result: Machine Vision Completes the Solution
The smart license plate solution was successfully tested internally and a pilot project is in progress for further testing. As our client extends their offering beyond physical components to complete digital solutions Dunn Solutions will continue to create additional value for their customers and users.