Motion-Aware Tiny Object Detection using YOLO-MoNet

Detecting moving objects is the process of recognizing and analyzing the actual movement of an object within a specified location. This task holds significant importance in computer vision, particularly in recognizing tiny moving objects from aerial images. Motion detection serves as a foundation in various applications such as object tracking and video surveillance algorithms. This paper proposed the YOLO-MoNet object detection model which is an improved version of YOLOv8. The proposed model focusing on the detection of motion-specific tiny objects across diverse visual environments. We selected the PESMOD dataset for experiments, which comprises challenging, high-resolution images featuring tiny objects across various sequences, distinguishing it from other datasets. The proposed YOLO-MoNet model outperforms other state-of-the-art algorithms across all six classes. Experimental results show that the YOLO-MoNet model achieved the highest Mean Average Precision (mAP) of 99%, Precision of 94%, and Recall of 100% for the POR class. Across all six classes, the mAP remains consistently high at 84%, resulting in an F1 score of 0.81. Furthermore, the cls loss for detecting tiny objects was significantly reduced by 80%.