Basic Information
- Project Title: Automatic MTO Generation from Engineering Drawings
- Name: Nutan Choudhary
- Project: SRE
- Semester(s): 7
- Guide: Prof. Siddharthdutta Gupta
Abstract
This industry project involved direct collaboration with a company, where we were provided engineering drawings of instruments represented by specific symbols and their corresponding costs. Our main objective was to develop an algorithm capable of accurately detecting the components in these images, matching them with the cost database, and creating a comprehensive bill of materials. However, the task proved to be more challenging than expected due to the complexity of the images. The instrument images were quite intricate, containing numerous tiny parts combined into a single PDF image. Achieving precise component detection with high accuracy became the primary challenge. Many of the components had minute differences, making it even more demanding. To improve the accuracy, we continuously optimized the algorithm using various image processing techniques. The methodology involved a multi-step process, starting with text detection and recognition using Optical Character Recognition (OCR) algorithms. Line identification was performed using the Hough Transform, and instruments and symbols were detected by tracing object edges and performing template matching. Label association was established by matching objects with templates extracted from legend sheets.
Any courses you completed relevant to the project
Image processing, Intro to ML
Describe your experience on the project
I previously worked on an OCR project in the healthcare domain, where we focused on detecting simple handwritten text from images and converting it into digital text. The handwriting belonged to doctors or medical staff. When I started this new project, which involved symbols and digital text, I initially thought it would be relatively straightforward. However, I quickly realized that it was more coding-intensive and required continuous optimization of the algorithms and trying new techniques. Unlike other projects where you can find existing code on platforms like GitHub, in this case, the symbols were delicate and unique, making it necessary to train our model from scratch. The main challenge was achieving high accuracy in symbol detection.
The workload wasn’t overwhelming, especially if you had prior experience in this domain. However, for those new to the field, you need to put in significant efforts. The potential for making a significant contribution and achieving publication in this less-explored domain is promising with sincere efforts.
Describe your experience with the guide
In this project, our team consisted of 3-4 members, including PhDs, Mtech, and Btech. We had the opportunity to work directly with the company, and while they did not insist, we decided to maintain biweekly reporting to keep them informed about our progress. I had only three meetings with my guide throughout the semester because we were directly communicating with the company. However, the professor was always available and easily accessible in his office, allowing us to meet him whenever necessary.
Our guide gave us the liberty to choose our own topic and work in our preferred manner. He rarely interrupted our work, allowing us to take breaks and compensate for them later. So we had enough time for midsem and endsem preparations. I think even slight progress made in the project could lead to an assured AA grade.
Any advice for anyone considering a project under the same guide? Any other professors working in similar fields?
He is one of the very chill professors and provides students with the liberty to work at their preferred speed, without any pressure to meet specific goals or achievements.
Prof. Ganesh Ramakrishna is also working on OCR.