摘要
近几年频发的踩踏事故,使得商业领域、交通管制领域和公共安全领域对人流统计和预测的需求逐渐上升.但是当前的人流统计手段往往采用传统的机械式或人力统计方法,不仅效率低,干扰正常的人员行走以及人流速度,还不能满足实时性统计需求.提出的方法采用基于深度学习的目标检测和目标追踪算法,能够对人流量进行实时的统计,并且通过标定和坐标转换能够生成人流热力图,方便实时监控.从摄像头数据采集到最终的人流量预测与预警,整个过程形成完整的态势感知系统,功能稳定,满足实时性需求.
Stampede incidents were on the rise in recent years which have led to a gradual increase in demand for flow statistics and forecasts. However, the current people flow counting methods often use traditional mechanical or manual statistical methods to do the job, which not only have low efficiency, interfere with pedestrian walking and flow speed, but also cannot meet real-time requirements. The method proposed in this paper adopts deep learning-based object detection and object tracking algorithm, which can calculate people flow speed in real time. It could also generate people flow heat map for real-time monitoring through calibration and coordinate transformation. From the camera data collection to the final people flow prediction and early warning, the whole process forms a complete situation awareness system which has stable functions and meets real-time requirements.
作者
李敏
杨阳
王钤
孟博
李凌寒
白入文
杜虹
Li Min;Yang Yang;Wang Qian;Meng Bo;Li Linghan;Bai Ruwen;Du Hong(School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044;Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100093;School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100093;School of Optics and Photonics,Beijing Institute of Technology,Beijing 100081)
出处
《信息安全研究》
2019年第6期488-494,共7页
Journal of Information Security Research
关键词
人流量统计
态势感知
目标检测
目标追踪
深度学习
视频监控
people flow counting
situation awareness
object detection
object tracking
deep learning
video surveillance
