Enhance PV Panel Detection Using Drone Equipped With RTK

With the exponential growth of solar energy, Photovoltaic (PV) installations have exceeded 500GW globally. At this scale, autonomous PV plant inspection solutions utilizing Unmanned Aerial Vehicle (UAV) can offer critical services to ensure early fault detection. The detection of the PV panel is an important part since, without the detection, we can’t perform an automatic inspection on the PV panels. Therefore, researchers focused on the PV panel detection through the utilization of image processing filters on a single image, stitched image, or from a video frame.

For this study, we program the UAV to follow certain predefined waypoints (GPS coordinates) to fly over the PV panels. The UAV flight path and height varies for each experiment. The tested heights are 16, 18, and 20 meters. Also, two flight paths were selected for the UAV during the experiment. The flight path depends on the waypoint selected, where the UAV will follow the predefined waypoints. The UAV will start recording the data from the first waypoint and will stop recording at the final waypoint. Two different data types are collected which are video file (thermal & RGB) and UAV coordinates from the Real-Time Kinematic (RTK). The collected data are post-processed through several layers of filters for Image processing, where the panel will be detected and labeled with ascending sequence order. The collected video frames will go through several layers of filters, which are filter to remove the noise and sun glare from the frame, followed by a black and white filter to reduce the computational complexity, then a blur filter to assist on the canny edge detection algorithm, which is then applied to the Canny edge algorithm to detect the edges from the frame, followed by contour algorithm, and finally convex hull algorithm is used to draw a bounding box around the PV panels.

After that, the UAV coordinates will be integrated with the image processing results to ensure unrepeated labels, through the usage of data association. There are many types of data association, which are Global Nearest Neighbor (GNN), Hungarian algorithm, and k-best assignment algorithm. This paper will ensure more robust PV panel detection since it is utilizing two different input sources and using the GNN to ensure unrepeated PV panel detection. The novelty of this work is the proposed algorithm, which utilizes the RTK and video data to ensure unrepeated PV panels detection from each frame. The proposed algorithm can be used for live PV panel detection and can ensure a more successful and unrepeated detection.