Resumen de UAV vision system: Application in electric line following and 3D reconstruction of associated terrain
In this work, a set of vision techniques applied to a UAV (Unmanned Aerial Vehicle) images is presented. The techniques are used to detect electrical lines and towers which are used in vision based navigation and for 3D associated terrain reconstruction. The developed work aims to be a previous stage for autonomous electrical infrastructure inspection. This work is divided in four stages: power line detection, transmission tower detection, UAV navigation and 3D reconstruction of associated terrain. In the first stage, a study of algorithms for line detection was performed. After that, an algorithm for line detection called CBS (Circle Based Search) which presented good results with azimuthal images was developed. This method offers a shorter response time in comparison with the Hough transform and the LSD (Line Segment Detector) algorithm, and a similar response to EDLines which is one of the fastest and most trustful algorithms for line detection. Given that most of the works related with line detection are focused in straight lines, an algorithm for catenary detection based on a concatenation process was developed. This algorithm was validated using real power line inspection images with catenaries. Additionally, in this work a tower detection method based on a feature descriptor with the capacity of detecting towers in times close to 100 ms was developed. Navigation over power lines by using UAVs requires a lot of tests because of the risk of failures and accidents. For this reason, a virtual environment for real time UAV simulation of visual navigation was developed by using ROS (Robot Operative System), which is open source. An onboard visual navigation system for UAV was also developed. This system allows the UAV to navigate following a power line in real sceneries by using the developed techniques. In the last part a 3D tower reconstruction that uses images obtained with UAVs is presented.}, keywordenglish = {line detection, inspection, navigation, tower detection, onboard vision system, UAV.
