Robust Object Detection in Jam-Packed Scenes Through Enhanced YOLOv8: A Deep Learning Method

Dimension variation, obstruction, and occlusion pose significant difficulties to object detection in heavily surrounded situations. The You Only Look Once (YOLO) v8 method is improved in this study by adding an optimized loss function, enhancing accuracy and robustness in such challenging circumstances. The model incorporates a number of breakthroughs, including a method that uses attention to concentrate on pertinent regions, a multi scale feature fusion technique to manage distinct object sizes, and a more advanced backbone network with new convolutional layers for improved extraction of features. The primary new feature is an improved loss algorithm that is intended to more successfully handle localization issues and class imbalance. This loss function uses metrics based on Intersection over Union (IoU) to improve localization accuracy and focal loss to address class imbalance. Full-blown tests are carried out on benchmark datasets such as Pascal VOC and COCO, in addition to a custom-designed dataset intended for dense situations. When compared to other cutting-edge object identification models and the baseline YOLO v8, the improved YOLO v8 performs better. Significant gains in mean Average Precision (mAP) and recall measures are demonstrated by the quantitative data. The suggested improvements to YOLO v8 maintain real-time processing capabilities while pushing the limits of object recognition in clogged environments, making it suitable for real-world applications in surveillance, autonomous driving, and urban scene analysis.