Ecomess OCR Reader

Challenge

Ecomess approached us with a request to develop a semi-automated OCR system for reading values on newly manufactured water meters after undergoing standard tests in the factory. The key individuals involved in the project were the CEO and the Project Manager. After experiencing some challenges while working with freelancers, our Partner made the decision to present their idea to us, and we took on the task.

The goal was to improve the production process by eliminating the manual step of transcribing flow values and instead using software to read and store the values electronically. 

The system eliminates the manual step of transcribing flow values from the water meter after validation and recording them. It achieves this by utilizing software equipped with a custom neural network that reads the value from the meter dial and saves it in the memory of the water meter as the initial state, all with a single click from the operator.

Going more into detail, the goal of this system was to install it in their factory in a way that it would read values achieved on newly manufactured water meters after undergoing a series of tests. The installation would involve placing a camera connected to a computer at a specific stage on their production line. As the water meter moved on the conveyor belt, the camera would capture an image and send it to the computer where the system was installed. The system would then analyze the image using OCR technology to read the desired values. It was designed to be only partially automated. After the analysis and value extraction, the result would be displayed on the screen, where an employee would manually confirm or correct it if any discrepancies occurred. This project was motivated by the fact that OCR solutions are not flawless but our Partner wanted an accurate representation of the values on the water meters within the system.

Project's scope

To develop a suitable solution, we conducted a workshop with Ecomess, during which we identified the challenges and specific requirements, and gained a comprehensive understanding of the production process.

The project began with the preparation of training datasets, which involved taking photos of the meter displays in various numerical configurations. Next, our developer spent several weeks developing the program and training the neural network. The training process required significant computational power and was conducted on a separate PC due to its resource-intensive nature.

After the development phase, we created a prototype of the system. It performed well, accurately recognizing values in approximately 90% of cases. We added a user interface (UI) to this software to make it easier to use and delivered the software bundle, along with installation instructions, to our Partner.

Our team

• Python developer
• Project Manager
• QA tester