Image generation for efficient neural network training in autonomous drone racing
Autonomous drone racing faces challenges with traditional gate detection due to varying conditions. This work proposes a semi-synthetic dataset combining real backgrounds and 3D renders for training convolutional neural networks for gate detection.