Long Range Object-Level Monocular Depth Estimation for UAVs

Computer vision-based object detection is a key modality for advanced Detect-And-Avoid systems that allow autonomous flight missions of UAVs. While standard frameworks do not predict the actual depth an object, this information crucial to avoid collisions. In paper, we propose several novel extensions state-of-the-art methods monocular from images at long range. Firstly, Sigmoid and ReLU-like encodings when modeling estimation as regression task. Secondly, frame classification problem introduce Soft-Argmax function in calculation training loss. The are exemplarily applied YOLOX framework. We evaluate performance using Amazon Airborne Object Tracking dataset. addition, Fitness score new metric jointly assesses both performance. Our results show proposed outperform approaches w.r.t. existing, well metrics.