Inhaltszusammenfassung

This work presents different methods and algorithms for the application of Image Assisted Total Stations (IATS). IATSs combine the accuracy of the angle measurement of a total station with photogrammetry and the associated possibility of image processing and analysis. The integration of cameras makes it possible to capture distinctive points on the object without the need for signalling. This offers the advantage that no access to the object is required, as distinctive points can be observed ...This work presents different methods and algorithms for the application of Image Assisted Total Stations (IATS). IATSs combine the accuracy of the angle measurement of a total station with photogrammetry and the associated possibility of image processing and analysis. The integration of cameras makes it possible to capture distinctive points on the object without the need for signalling. This offers the advantage that no access to the object is required, as distinctive points can be observed directly and without the use of geodetic targets such as prisms. By using methods such as template-based or feature-based matching, points are extracted and their movement detected. A disadvantage is the dependency on sufficient ambient lighting or illumination of the point to be observed. First, the developments of various imaging total stations since the year 2000 are presented. A distinction is made between developments by research institutions and commercial versions. The various hardware components used as well as the image processing algorithms are presented in more detail. The focus is on the application in the field of Structural Health Monitoring (SHM). In particular, the photogrammetric methods for determining and tracking unsignalised points are discussed. Due to changing light conditions and atmospheric influences, automation is not trivial. Since deformation measurements are usually performed repeatedly, methods for the detection of discrete points are of particular interest. It is shown that the MoDiTa (Modular Digital Imaging Total Station) measurement system can be successfully used to determine deformations and dynamic characteristics. Using various measurement examples on bridge structures, it is shown how the vertical deflection and the vibration behaviour can be successfully recorded and how they contribute to the assessment of the structure condition. In the following, a solution approach for the self-calibration of a MoDiTa is presented. This approach extends an existing Total Station modularly by a digital camera and enables calibration directly on site. This extension opens up further fields of application such as high-frequency target acquisition or tracking of moving targets. Calibration is performed using image processing and equalisation methods and is largely automated. The crosshair plane is captured in each image and thus provides identical points in both the camera image and the reference image. Due to the modular design and the associated non-exactly repeatable mounting of the camera, and in order to exclude any possible movement of the camera, the crosshair is continuously observed during the measurement. The methods and image processing algorithms required for this are explained in detail. Using an exemplary self-calibration, the parameters of the calculation are presented and it is shown that they also retain their validity for different distances to the object or focus positions. In the third section, the use of artificial intelligence in combination with IATSs is considered. For narrowly defined tasks, machine learning methods, in this case deep learning, have proven to be an effective tool. The necessary data sets were self-generated and semi-automatically annotated for the essential training. The present approaches are limited to classification and object detection by means of supervised learning. Two different use cases are explained. The first approach is to minimise the manual intervention of the operator by classifying the captured images from the MoDiTa measurement system. The developed control software is supposed to check the captured image for further use, which is necessary for the crosshair recognition. In order to determine the crosshair using classical image processing algorithms such as edge detection, an image must be used that is sufficiently exposed and free of disturbances. The crosshair should not be detected in this step, but only the background should be evaluated to see if it is appropriate for further use. So far, this decision is made by the user and offers potential for automation. The second use case involves finding geodetic targets in captured images using IATS and Deep Learning. It examines whether a supposed target (retroreflective and non-retroreflective) can be seen in the captured image and identifies and locates it (object detection). By means of a so-called bounding box, the target found is located in the image and the rough direction to the target in the superior system can be determined via the pixel information supplied. Already implemented applications for target identification are to be supported in this way and extended with additional information. The various methods and algorithms presented are described in detail and evaluated for their practicality. The respective strengths and weaknesses are discussed and it is shown that IATSs have great potential, but that this potential is not being fully exploited at the present time. Especially the area of image processing and analysis has been neglected so far or the already existing product range of possible methods and algorithms has not been fully utilised.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
Geophysik und Geodäsie

DDC Sachgruppe:
Ingenieurwissenschaften