测绘通报 ›› 2025, Vol. 0 ›› Issue (9): 26-33.doi: 10.13474/j.cnki.11-2246.2025.0905

• 生态环境动态监测 • 上一篇    下一篇

基于双目视觉的林草火灾检测与测距方法

朱元彩1,2, 孙立瑛3, 张凡3, 吴兆立1,4, 高向东3, 金雷1, 李晓东5, 李若瑜1,2   

  1. 1. 江苏建筑职业技术学院, 江苏 徐州 221116;
    2. 徐州市水资源高效利用与生态安全应用技术工程研究中心, 江苏 徐州 221116;
    3. 天津城建大学控制与机械工程学院, 天津 300384;
    4. 江苏省智能视觉识别与数据挖掘工程研究中心, 江苏 徐州 221116;
    5. 新疆应用职业技术学院, 新疆 奎屯 833200
  • 收稿日期:2025-02-05 发布日期:2025-09-29
  • 作者简介:朱元彩(1972—),男,硕士,正高级工程师,研究方向为计算机技术应用、视觉识别。E-mail:zxcvpoiu@126.com
  • 基金资助:
    江苏省电子信息职业教育研究重点课题(JSDX2021-I09);新疆维吾尔自治区产学合作协同育人项目(258);江苏省建设系统科技项目(2022ZD048);徐州市2023年度产学研专题对接活动落地项目(B类)(KC23403)

Fire detection and ranging method based on binocular vision

ZHU Yuancai1,2, SUN Liying3, ZHANG Fan3, WU Zhaoli1,4, GAO Xiangdong3, JIN Lei1, LI Xiaodong5, LI Ruoyu1,2   

  1. 1. School of Electronics and Information Engineering, Jiangsu Vocational Institute of Architectural Technology, Xuzhou 221116, China;
    2. Xuzhou Water Resources Efficient Utilization and Ecological Security Engineering Research Center, Xuzhou 221116, China;
    3. School of Control and Mechanical Engineering, Tianjin Chengjian University, Tianjin 300384, China;
    4. Jiangsu Province Engineering Research Center of Intelligent Visual Recognition and Data Mining, Xuzhou 221116, China;
    5. Xinjiang Career Technical College, Kuitun 833200, China
  • Received:2025-02-05 Published:2025-09-29

摘要: 林草火灾已成为严重的公共安全问题,经常导致重大人员伤亡和财产损失。本文提出了一种目标检测与立体匹配相结合的火灾目标测距方法,实现对火灾目标的识别和定位。采用3层SGBM算法与YOLOv5目标检测相结合,进行WLS滤波;通过改进3层SGBM算法,对目标深度信息进行不同尺度的优化,提高了大尺度深度图像的准确性和稳定性,提出了一种基于正三角形的火灾目标检测框中心点距离计算方法,剔除无效点,提高测距精度。试验结果表明,测试距离中检测目标占比在4%以内时,改进的SGBM算法能够达到较高的精度,相对定位误差在2%以内。本文方法具有较高的精度和可靠性,可以应用于林草火灾监测预警和灭火救援等领域。

关键词: 林草火灾, 火灾测距, 双目视觉, 目标检测, 立体匹配

Abstract: Forest and grassland fires have become a serious public safety issue in today's society,often leading to significant casualties and property losses.This paper proposes a fire target ranging method that combines object detection with stereo matching to achieve the identification and localization of fire targets.The method utilizes a three-layer SGBM algorithm in conjunction with YOLOv5 object detection,applying WLS filtering and optimizing the depth information of targets at different scales to enhance the accuracy and stability of large-scale depth images.Experimental results show the system has high accuracy.When the detected target proportion in testing distance is within 4%,the improved SGBM algorithm achieves high precision,with the relative positioning error kept within 2%.The proposed method exhibits high accuracy and reliability,making it applicable to fire monitoring,early warning,firefighting,and rescue operations in forest and grassland areas.

Key words: forest and grassland fires, fire ranging, binocular vision, object detection, stereo matching

中图分类号: