Landing Pad Recognition Technology Based on Semantic Image Segmentation Using Virtual Reality Dataset

We explored the recognition of a landing pad by using semantic image segmentation with the integration of a virtual reality dataset. To create a virtual 3D environment similar to the real environment, we applied this technology to create a 3D map based on aerial photos. In the resulting virtual reality, we created a scenario in which a drone equipped with a camera acquired images of the landing pad. The program was designed to place a landing pad in a virtual reality, acquire images of the landing pad using a camera in virtual reality, and automatically perform image segmentation labeling for the landing pad. An experiment was conducted in which a deep neural network recognized the landing pad from the images captured by an actual drone. The network used only the dataset constructed with automatic labeling technology in virtual reality. The experimental results confirmed that the actual landing pad image could be recognized even by using only the dataset collected in virtual reality. © ICROS 2024.