This master thesis proposes a deep learning method combined with a heuristics approach to automatically extract dimensions from bridge engineering drawings. It is part of the TwinGen research project, whose goal is to develop methods that automatically generate digital twins of existing infrastructure structures. The proposed solution consists of an image classification algorithm that classifies the image by type of view, then passes it to an object detection algorithm that detects the numbers of the important dimensions on the image and then is followed up by a heuristics algorithm that matched each detected value to its corresponding dimension. The convolutional neural network architecture MobileNet is used for both classification and detection. Results show that the classification task can be performed reliably, with an 85.71% validation accuracy rate, while the detection task is performing with a moderate success rate.     «

This master thesis proposes a deep learning method combined with a heuristics approach to automatically extract dimensions from bridge engineering drawings. It is part of the TwinGen research project, whose goal is to develop methods that automatically generate digital twins of existing infrastructure structures. The proposed solution consists of an image classification algorithm that classifies the image by type of view, then passes it to an object detection algorithm that detects the numbers o...     »