Motion Feature Aggregation for Video Object Detection Using YOLO Approaches

This paper addresses motion feature aggregation for video object detection using YOLO approaches and intelligent transportation systems (ITS). The rapid and accurate vehicle detection and identification is challenging due to interference aspects of images or video frames showing automobiles. In this paper, we present advanced YOLOv7-A, an optimized version of the YOLOv7 model for object detection and classification, to address this issue. The suggested model makes a technique for paying close attention in order to hide picture interference in any dimension. Furthermore, YOLOv7 employs a tweaked version of the Path Aggregation Network's (PAN) Feature Pyramid Network (FPN) to improve the quality of image. As a result, the model's object identification and classification performance are enhanced, and the items may be placed more reliably in 3D space. We considered different datasets for experimental results based on the suggested YOLOv7-A model that obtained good performance on the basis of precision and F1 score than the YOLOv7 model and also took comparison using two other methods such as Faster R-CNN and EfficientDet with different performance values.