VISION-BASED AUTONOMOUS INSPECTION OF VERTICAL STRUCTURES USING UNMANNED AERIAL VEHICLE (UAV)

High-rise structures like a chimney, flare stacks, storage tanks, cooling towers, electric line poles, and communication towers become a vital part of any industry and are very common in day-to-day life. These vertical structures required a proper and frequent inspection to run the industries safely and stay profitable. The inspection of the industrial structure is done in several stages, and vision-based inspection is the very initial, oldest, and simplest method to detect and locate the surface defects and anomalies. The existing traditional methods of vision-based inspections are unsafe, time-consuming, and are an extra financial burden on a company and the existing robotics inspections methods are ineffective, slow, and exhausting due to complex dynamics, structure, and weights. In this research work, we are proposing a fully autonomous visual inspection approach to inspect the vertical structure using a quadcopter. The unmanned aerial vehicle (UAV) equipped with cameras and non-contact sensors is simulated with the help of robot operating systems (ROS) and a visualization tool gazebo. To examine the developed algorithms, a simple black-colored cylindrical vertical structure is prepared in the gazebo. Here, the UAV first detects and locates the vertical structure in the image frame using a classical computer vision algorithm and will extract some desired features. The feature information coming out from the image frame will be fed into the heuristically tuned control algorithms for navigating and positioning the UAV around the vertical structure. For this work, offset control and width/radius of rotation control algorithms have been developed for positioning and trajectory tracking. Due to the image frame localization and positioning, the GPS dependency is not there, and it can operate in GPS denied the environment also. The simulation results are quite satisfying, and the overall performance of the computer vision algorithms and control algorithms is satisfactory.