Inhaltszusammenfassung

Surveying Engineers use a range of measurement techniques to determine the precise location of objects. The tasks and disciplines as well as the methods used offer a more comprehensive portfolio than ever before. In addition to conventional measurements, image-based sensor technology is increasingly being used to generate an expanded information spectrum of data. Suppliers of various image data are e.g. UAV (Unmanned Aerial Vehicle) or camera-based total station. By automating the evaluation ...Surveying Engineers use a range of measurement techniques to determine the precise location of objects. The tasks and disciplines as well as the methods used offer a more comprehensive portfolio than ever before. In addition to conventional measurements, image-based sensor technology is increasingly being used to generate an expanded information spectrum of data. Suppliers of various image data are e.g. UAV (Unmanned Aerial Vehicle) or camera-based total station. By automating the evaluation processes, the images create advantages and support humans at the same time. This is driven by the use of artificial intelligence, especially deep learning (DL). DL has been experiencing an upswing for several years, especially in the area of image processing of large amounts of data. The focus is on pattern recognition in 2D data sets using artificial neural networks. The purpose of this study is to merge the two subfields and thereby enable automatic, image-based pattern recognition using DL methods in a geodetic context. We will investigate the different uses and issues as well as evaluate the approaches in practice. This study presents two approaches in three different application examples. The sensor technology of the image captures as well as the searched patterns differ among each other. We apply the methods classification and object detection in the branch of supervised learning. The first approach focuses on the automatic search of different target designs. First, we address the research question as to what extent the potential of image acquisition by means of a camera-based total station can be used in combination with DL. As targets, we use a classical universal-prism, a 360° prism, a black/white target and a solid sphere. Second, we investigate uniquely identifiable targets from aerial imagery, with image data obtained from UAV flights. We exclusively use a black and white target design, which resembles a star pattern over eight lines crossing in the middle. No data fusion takes place under the two application examples because the consideration of the questions is different. In order to increase the datasets, we use data augmentation techniques. In addition to commercial software, we use open-source software libraries as well as pre-trained. The safeguarding and monitoring of construction sites is another area of expertise for surveyors. In the second approach, we are testing the automatic detection of trains by using image-based sensor technology to enable the construction site personnel to clear the track area in time. The results show a great potential of application-based DL in the field of image-based pattern recognition. The sensor technology and the delivered image data show an efficient use.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
Informatik

DDC Sachgruppe:
Ingenieurwissenschaften