Scholarlink

Surveillance plays a critical role in monitoring people and vehicles in various settings. This study introduces CrowdISEntrY, a novel approach utilizing YOLOv8 and DeepSort for precise object counting in both CCTV and drone videos. YOLOv8, leveraging a feature pyramid network, ensures accurate object detection, followed by DeepSORT for object tracking using Kalman filters and matching cascade. This integrated approach facilitates effective entry and exit counting, thereby enhancing surveillance across diverse applications. Comparative analysis with YOLOv3, YOLOv5, YOLOv7, and YOLOv8, along with DeepSort, demonstrates superior performance. In CCTV videos, CrowdISEntrY achieves a precision of 0.91, recall of 0.82, F1-score of 0.86, and mAP of 0.72. In drone videos, it outperforms other models with a precision of 0.88, recall of 0.77, F1-score of 0.82, and mAP of 0.66. Overall, CrowdISEntrY exhibits an accuracy of 98.07% for entry and exit counting, surpassing existing models and demonstrating its efficacy in real-time surveillance scenarios.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CrowdISEntrY: Intelligent Object Entry-Exit Counting Framework and Comparative Analysis in CCTV and Drone Surveillance Systems

  • 章节

  • 发表日期：

    2026年

查看全文
  • Siddharth Shankar Das,
  • A. Ancy Micheal,
  • Ambar Kumar,
  • Moin Ahmed Zahir

摘要

Surveillance plays a critical role in monitoring people and vehicles in various settings. This study introduces CrowdISEntrY, a novel approach utilizing YOLOv8 and DeepSort for precise object counting in both CCTV and drone videos. YOLOv8, leveraging a feature pyramid network, ensures accurate object detection, followed by DeepSORT for object tracking using Kalman filters and matching cascade. This integrated approach facilitates effective entry and exit counting, thereby enhancing surveillance across diverse applications. Comparative analysis with YOLOv3, YOLOv5, YOLOv7, and YOLOv8, along with DeepSort, demonstrates superior performance. In CCTV videos, CrowdISEntrY achieves a precision of 0.91, recall of 0.82, F1-score of 0.86, and mAP of 0.72. In drone videos, it outperforms other models with a precision of 0.88, recall of 0.77, F1-score of 0.82, and mAP of 0.66. Overall, CrowdISEntrY exhibits an accuracy of 98.07% for entry and exit counting, surpassing existing models and demonstrating its efficacy in real-time surveillance scenarios.