The accurate detection and tracking of objects in drone imagery is still an open problem because of the frequent rotation of the drone during the flight. These rotations affect greatly the orientation of the captured images of the drone. Traditional object detection systems like YOLOv5 produce bounding boxes that are not aligned with the rotated objects, therefore, suboptimal thumbnail generation results which affect the subsequent analysis tasks. This paper presents the Sha Rotation Thumbnail (SRT) algorithm, a new postprocessing technique, which, by evaluating from various perspectives, step-wise agrees with the most suitable object representation. The new method first implements different rotations (0°, 90°, 180°, and 270°) of the detected objects and then selects the thumbnail configuration that is most consistent with the object’s geometrical characteristics. A dataset of 7,425 drone captured images has been used to check the performance of the proposed system compared to a basic YOLOv5, resulting to SRT improved system giving a 9.5% increase in F1score (0.896 vs 0.818) and substantial gains in both precision (0.919 vs 0.844) and recall (0.875 vs 0.793). Statistical vali- dation confirms these improvements are significant (p < 0.001) across all performance metrics. The algorithm’s efficient implementation main- tains real-time processing capabilities while resolving a critical limitation in current aerial object detection systems.