**6. Conclusions**

The presented results allow us to conclude that the use of laser scanning to obtain reference trajectories regardless of the availability of the GNSS signal (which often disappears in urban environments) is effective, and it can be applied as part of the methodology for testing vision-based localization systems, as the accuracy obtained is sufficient.

This research has shown that the trajectories estimated by a state-of-the-art visual SLAM system are of acceptable accuracy in the context of local navigation in the urban environment. The obtained accuracy of the vehicle pose estimates is also sufficient for localization of selected objects on the roadsides—such as billboards considered in the CityBrands project.

The multi-loop paths of the presented experiments, which are not present in the publicly available datasets for SLAM, for example, KITTI [25] made it possible to demonstrate the crucial role of topological loop closing in obtaining globally consistent trajectories. A clear decrease of this global consistency was observed for the trajectories obtained from the images acquired using smartphone with its rolling-shutter camera. This can be attributed to the smaller number of point features detected in the lower-quality images that in some cases were insufficient to detect a loop using the bag-of-visual-words approach. This was observed for trajectories that otherwise had local errors not greater than their counterparts estimated from images collected by a professional global-shutter camera. However, supplementing the SLAM algorithm with only locally available (at selected points) GNSS data (which here served only as ground truth) should allow to removal of the observed shortcomings. In general, this research demonstrated that even low-cost commodity hardware can be used to obtain useful trajectories of a vehicle in urban environment if recent algorithms are applied for data processing, and the entire processing pipeline is implemented carefully, focusing on proper calibration of the sensors.
