Scholarlink

With the popularity of Closed Circuit Televison (CCTV) camera systems, the need for fire detection systems using CCTV cameras has increased continuously. Conventional thermal and smoke sensors work well to detect nearby and indoor fires but cannot warn about remote situations. This paper proposes a camera-based fire detection method using a combination of a preliminary detector and a Convolutional Neural Network (CNN) classifier. The preliminary detector detects bounding boxes using primary image features such as color, brightness, and motion features. Based on these bounding boxes, we build and train a CNN to eliminate false positives and get an accurate fire detection result. The proposed method is evaluated on seven fire videos under various conditions, reaching an average accuracy of 96.58% and over 90% in experimental setups.

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

Fire Detection in Surveillance Camera Systems Using a Deep-Learning-Based Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Quang Thi Nguyen,
  • Minh Kha Pham,
  • Khanh Toan Vu,
  • Huu Hung Nguyen,
  • Hien Ha Thi

摘要

With the popularity of Closed Circuit Televison (CCTV) camera systems, the need for fire detection systems using CCTV cameras has increased continuously. Conventional thermal and smoke sensors work well to detect nearby and indoor fires but cannot warn about remote situations. This paper proposes a camera-based fire detection method using a combination of a preliminary detector and a Convolutional Neural Network (CNN) classifier. The preliminary detector detects bounding boxes using primary image features such as color, brightness, and motion features. Based on these bounding boxes, we build and train a CNN to eliminate false positives and get an accurate fire detection result. The proposed method is evaluated on seven fire videos under various conditions, reaching an average accuracy of 96.58% and over 90% in experimental setups.