Addressing the challenges of complex backgrounds, small parabolic targets, indistinct appearance features of parabolas, and easy loss of parabolic tracking in high-altitude parabolic object detection, this paper proposes a hybrid approach integrating the Mixture of Gaussians Background Modeling (MOG) algorithm with neural networks and an improved Simple Online and Realtime Tracking (SORT) algorithm for parabolic tracking. Firstly, to mitigate the issues of small target parabolas and complex backgrounds, a region-specific conditional filtering is introduced to reduce non-parabolic foreground in foreground detection while preserving parabolic foreground. Secondly, to tackle the problem of indistinct appearance features of parabolas, a multi-frame channel fusion technique is employed to enhance motion features, and a lightweight classification network is designed to differentiate parabolic objects. Finally, to address the challenge of easy loss of parabolic tracking, the state space and matching metrics of SORT are improved to better match parabolic trajectories. Experimental results demonstrate that the improved parabolic detection method reduces detection quantity by 97\% compared to the original MOG algorithm while exhibiting a 7\% decrease in recall rate. Additionally, compared to the original SORT algorithm, the improved parabolic tracking method reduces the number of ID switches by 50%, increases the MOTA metric by 8%, and increases the TIOU metric by 7%.
Code: https://figshare.com/articles/dataset/High-Altitude Object Detection algorithm/25778430. Dataset:https://figshare.com/articles/dataset/High-Altitude Object Detection Dataset/25778289.