Imagine an application that gives vehicles access through a gate by identifying their plate numbers with AI-powered technology. Students of TUMO Labs, an EU-funded science, technology, and engineering program, took on the task with Yerevan-based web development company AOByte. With Vardan Girgoryants as project lead, students spent the past three months developing an optical recognition system that detects plate numbers.
“We succeeded in solving the problem in front of us,” said Vardan, a full stack developer at AOByte. “Of course, as a specialist I would have liked more time on the project, to help students experiment with different algorithms and eventually choose the best one. That would also be an interesting process.”
The ten project participants included students from several Armenian universities, most with a background in computer vision and machine learning, and young professionals taking their first steps in the field. Iren Okminyan, an AUA student who dreams of working in an AI and machine learning lab, participated in the project as her undergraduate thesis. “There were several projects to choose from as our final thesis,” explained Iren. “I was immediately drawn to the TUMO Labs project because it had a clear structure, timeline and objectives. You can say that through this experience I earned my diploma.”
Every week, first at TUMO Yerevan and then remotely on Zoom, participants convened for six hours worth of meetings and workshops. Despite the challenges of remote learning, students reacted to the change quite positively. “It was easy to switch to online learning because of the nature of the work and Vardan’s teaching style. I’ve also been more relaxed working from home because I don’t have to worry about running late,” said Lilit Manukyan, a student at Yerevan Polytechnic University.
The TUMO Labs project covered different types of machine learning, optimization algorithms, accuracy metrics, transfer learning, as well as neural networks, deep learning and Python libraries. On this foundation, students developed convolutional neural network models (CNN) to classify handwritten digits for MNIST, a database used for training image processing systems.
“Over the last three months, we’ve dived deep into the four subgroups of machine learning. The amount of information and pace of instruction have been ideal for me,” said Ani Gabrielyan, who studied finance in college and recently decided to switch careers to programming. Before joining the TUMO Labs project, Ani had taken online courses in AI and Machine Learning and was ready for a practical experience to apply her learning. “Now, I feel more confident in the field and applying for jobs,” she said.
By the end of the project, the group developed an application that retrieves a license plate number with scanning and outlining technology, then runs the characters through a database to, ultimately, grant a vehicle access through a gate…or not. As you can see, the gate access revolution is right around the corner! Watch this video by project participant Lilit Manukyan to learn more.