Predicting Space Debris Rotation Using the Debris Rotation Prediction Network (DRPN)

Tracking the rotation of space debris in Earth’s atmosphere is critical for predicting its trajectory and preventing collisions. While existing methods for detecting the velocity and direction of space debris are well-established, accurately predicting its rotation presents a unique challenge due to limited datasets and the complex behavior of debris. To address these challenges, we introduce the Debris Rotation Prediction Network (DRPN), a novel computer vision model that predicts the rotational motion of space debris using two frames from low-FPS video sequences. DRPN integrates convolutional neural networks (CNN) with custom keypoint matching mechanisms to overcome issues caused by environmental distortions, low resolution, and low frame rates. We developed a synthetic dataset using Unreal Engine to simulate various debris movements, including the effects of lighting distortions, and tested the model against traditional methods such as SIFT. Unlike traditional feature detection methods, DRPN can adapt to space environments with low-FPS data, providing high accuracy in rotational vector predictions. Experimental results show that DRPN significantly outperforms existing techniques, achieving lower rotational vector error and angular velocity error. Our findings suggest potential for further accuracy improvements by incorporating additional video frames and real-world data into the model. The proposed approach holds promise for real-world deployment in space debris tracking systems.