MASTERÂ Assignment
A comprehensive study on feature detection algorithms and fine tuning methods for efficient metadata extraction in construction drawing
Type : Master M-BIT
Period: Oct, 2023 - March, 2024
Student : Gupta, P. (Pratyush, Student M-CS)
Date Final project: March 18, 2024
Supervisors:
Abstract:
The thesis aims to investigate the potential of leveraging image processing algorithms to extract metadata from construction drawings, with the aim of enhancing efficiency and effectiveness in the industrial processes. The research begins by identifying existing challenges in business process workflows, such as lack of automation. Manual extraction of metadata from construction drawings can be time-consuming and error-prone, leading to inefficiencies in the workflow. Through a meticulous review of relevant literature and exploration of image processing methodologies, the study proposes an approach to automate metadata extraction from construction drawings. Central to the research are three key research questions: the feasibility of automating metadata extraction, the identification of optimal detection techniques, and the development of approaches to fine-tune algorithms for efficient results. By conducting experiments and evaluations, the study determines that certain image processing algorithms, when optimized through multiple fine-tuning techniques, can effectively extract metadata from construction drawings, even in scenarios with limited data availability. The findings of the research have significant implications for industries reliant on construction drawings, offering insights into cost-effective solutions for enhancing workflow efficiency and customer satisfaction. The optimization efforts resulted in significant achievement in macro-F1 and accuracy scores, with CCORR_NORMED achieving average scores of 0.81(macro-F1) and 0.83(Accuracy) for detection experiments conducted. Also, a proposed index, i.e. Detection Effectiveness Index (DEI) resulted in the score of 0.85 in case of CCOEFF_NORMED. These results highlight the potential of image processing algorithms in addressing real-world challenges, the feasibility of achieving meaningful outcomes with limited data. Finally, contributing to the advancement of knowledge and practices at the intersection of image processing, construction drawings, and industrial processes.