| Form of studies |
Bachelor |
| Title of the study programm |
Computer Science and Organizational Technologies |
| Title in original language |
Mašīnmācīšanās asistēta ugunsdzēšamo aparātu klasifikācija no attēliem |
| Title in English |
Deep Learning-Assisted Extraction of Fire Extinguisher Classification from Images |
| Department |
01B00 Riga Business School |
| Scientific advisor |
Inese Muzikante |
| Reviewer |
Pēteris Paikens |
| Abstract |
This thesis aims to contribute to the field of Scan to Building Information Modeling by developing a method for automatically extracting fire extinguisher classification information from images. Scan to BIM is a modern approach that generates accurate and detailed representations of buildings, which includes vital fire safety equipment information. Automating the extraction of fire extinguisher classification information can enhance the efficiency and accuracy of building inspections and emergency response planning.
The research focuses on the principles of computer vision and deep learning by exploring their potential applications in extracting fire extinguisher classification information from images. High-resolution images of fire extinguishers were collected in Riga, Latvia, and used to develop and test the proposed approach.
The research method consists of the following steps: data collection, preprocessing, and annotation of images; machine learning model selection and solution design; and evaluation of the method's performance. In this research YOLOv5 was utilized for region recognition, dlib library for landmark detection, image transformation techniques for improving text readability, and optical character recognition to extract text from the images.
The results demonstrate the potential of the proposed method in accurately identifying and extracting fire extinguisher classification information from images. This contributes to the creation of comprehensive and reliable Scan to BIM models, leading to improvements in fire safety inspection procedure. Despite the study's limitations, such as the use of a limited dataset and potential challenges in real-world applications, the findings suggest a promising direction for future research and improvements in automating text extraction within the Scan to BIM domain. Furthermore, the proposed approach demonstrates a general framework for text extraction from images with multiple objects.
Keywords: machine learning, computer vision, object recognition, optical character recognition, building information modeling, fire safety |
| Keywords |
Atslēgvārdi: mašīnmācīšanās, datorredze, objektu atpazīšana, optiskā rakstzīmju atpazīšana, būvniecības informācijas modelēšana, ugunsdrošība |
| Keywords in English |
Keywords: machine learning, computer vision, object recognition, optical character recognition, building information modeling, fire safety |
| Language |
eng |
| Year |
2023 |
| Date and time of uploading |
19.04.2023 22:25:04 |
