Accuracy and Real-Time Performance Evaluation of Pedestrian Detection Models Based on the COCO Dataset

Abstract

. The detection of pedestrians holds a very important position in application fields such as self-driving cars and intelligent monitoring systems. This research carries out a comparison between a traditional method of HOG+SVM and current deep learning models, including Faster R-CNN and YOLOv8, for the purpose of assessing their detection capability. The outcome of the experiments shows that HOG+SVM tends to produce failing detections in complex environments, while deep learning models demonstrate stronger robustness. Among them, YOLOv8s attains higher detection accuracy, whereas YOLOv8n provides faster processing speed.Under the existing experimental conditions, YOLOv8 achieves a reasonable balance between precision and computational efficiency. Faster R-CNN shows relatively stable performance but lower speed due to its two-stage structure. Evaluation metrics such as mAP and FPS are used for comparison. These results provide practical guidance for selecting suitable pedestrian detection models in real-time applications.