Examenspresentation av Oskar More Arvidsson

Examiner: Thomas Rylander

Abstract:
Radio waves at higher frequencies in the current and future generations of radio networks are more sensitive to details in the environment as they propagate. Previous measurements have shown that street furniture such as poles and trees can have a non-negligible effect on the characteristics of the radio propagation. Street poles can contribute significantly in around street corner situations to path gain, in particular at higher frequencies. Furthermore, the scattering contributes to the richness of the channel. For site specific modeling, ray tracing simulations are needed. By including street poles in ray tracing simulations in this work, more realistic simulation results were obtained. This was seen through changes in Doppler frequency and path gain. This thesis focuses on positioning the poles in the site specific modeling, enabling inclusion of them in the ray tracing simulations. To begin with, street view panorama images including the poles in the interesting area were extracted. With deep learning algorithms, object detection was performed through panoptic segmentation and range detection through monocular depth estimation on the extracted images. Given the direction and distance output from images extracted at multiple camera positions, street poles along a road are positioned through triangulation and clustering with a positioning error of 3.5 meters. This is comparable to related approaches in the field. The errors are mostly due to limiting GPS accuracy for camera positioning and limitations of detecting distant poles.

Welcome!
Oskar and Thomas