摘要
为解决油田日常生产作业中缺乏危险作业区域的等级划分与自动识别方式以及缺乏人员踏入危险区域的识别方法。提出1种基于视频智能综合识别技术的全天油田危险区域入侵检测算法,该算法首先结合油田危险因素对油田危险区域进行危险等级的划分与危险区域的识别;然后,针对光照条件良好的白天场景,在训练数据集中融合油田作业区视频数据和公开行人数据集,弥补油田作业区入侵样本不足的问题,有效地增加模型的泛化性;针对光照条件差的黑夜场景,使用三帧差分法,背景减除法等算法对运动目标进行检测。研究结果表明:本文提出算法较YOLOv5方法的精度更高,在不同油田场景下精度可达91.83%,已在油田作业现场进行部署与应用。
In order to solve the lack of classification and automatic identification methods of dangerous operation areas and the lack of identification methods for personnel entering dangerous areas in daily production operation of oilfields,an intrusion detection algorithm for all-day oilfield dangerous areas based on the video intelligent comprehensive recognition technology was put forward.Firstly,the algorithm combined the oilfield risk factors to classify the oilfield dangerous areas and identify the dangerous areas.Then,for the daytime scenes with good lighting conditions,the video data of the oilfield operation area and the public pedestrian data set were integrated into the training datasets to make up for the lack of intrusion samples in the oilfield operation area and effectively increase the generalization of the model.For the dark night scenes with poor lighting conditions,the three-frame difference method and background subtraction method and other algorithms were used to detect the moving targets.The results showed that compared with the YOLOv5 method,the proposed algorithm had an accuracy of up to 90%in different oilfield scenes,and it had been deployed and applied on the oilfield operation site.
作者
田枫
白欣宇
刘芳
姜文文
于巾涛
TIAN Feng;BAI Xinyu;LIU Fang;JIANG Wenwen;YU Jintao(College of Computer and Information Technology,Northeast Petroleum University,Daqing Heilongjiang 163318,China)
出处
《中国安全生产科学技术》
CAS
CSCD
北大核心
2022年第3期68-75,共8页
Journal of Safety Science and Technology
基金
国家自然科学基金项目(61502094)
黑龙江省省属本科高校基本科研业务费项目(KYCXTD201903,2020YDL-11)
东北石油大学研究生教育创新工程项目(JYCX_11_2020)。
关键词
区域入侵
多目标跟踪
油田危险区域分级
智能分析
YOLOv5
area intrusion
multi-target tracking
oilfield dangerous area classification
intelligent analysis
YOLOv5