Dense Pedestrian Detection Based on Improved YOLOv8

In intelligent video surveillance systems, dense pedestrian detection technology is critical for comprehending and evaluating crowd dynamics. However, in complicated situations, detection tasks are difficult owing to occlusion by pedestrians and tiny target sizes. To increase detection accuracy, this study provides an updated YOLOv8 algorithm that integrates the CARAFE module, SE module, and Wise-IOU loss function. The experimental findings on the WiderPerson dataset suggest that the improved model ([email protected]) Up to 74.7%, in mAP at 0.5 At 0.95, it achieved 52%, which is 0.8% and 1.2% higher than the original YOLOv8n model, respectively. Improving the model on a more complex CrowdHuman dataset ([email protected]) Up to 73.5%, in mAP at 0.5 On 0.95, it reached 50.8%, up 1.0% and 1.3% respectively. These findings show that the suggested technique may greatly enhance dense pedestrian recognition performance, particularly when coping with occlusion and tiny targets, resulting in increased accuracy and resilience. This paper proposes a novel approach for the use of intelligent monitoring systems in dense crowd settings.