综合运动前景与运动行人的智能视频双重监测算法设计  

Design of intelligent video dual monitoring algorithm for integrated motion prospect and moving pedestrian

作　　者：刘彦琴[1] 李杰[1] 李嘉仁 LIU Yanqin;LI Jie;LI Jiaren(Xi’an Fanyi University,Xi’an 710105,China)

机构地区：[1]西安翻译学院,西安710105

出　　处：《自动化与仪器仪表》2025年第2期5-9,共5页Automation & Instrumentation

基　　金：陕西省文化和旅游厅课题(2023HZ1618)。

摘　　要：针对传统监控视频存在目标尺寸小、行人被遮挡导致检测精度和跟踪效果不佳的问题,提出一种综合运动前景与运动行人的智能视频双重监测算法。首先,基于YOLOv5算法,构建一个结合协同注意力机制的行人检测模型;然后以行人检测模型作为检测器,引入DeepSORT的行人跟踪算法,并在该算法基础上加入GhostNet-cot重识别网络和抗遮挡策略;最后将改进YOLOv5检测模型与增强数据关联的DeepSORT跟踪模型相结合,搭建一个综合运动前景与运动行人的智能视频双重监测模型,通过该模型实现运动行人准确检测和跟踪。实验结果表明,本模型的MOTA、HOTA、IDFI指标分别取值为48.86%、53.35%和56.66%,均高于传统的DeepSort模型、GAN-RNN模型和ECA-BIFPN模型。且本模型的IDSW和参数量分别为204和9.2 M,明显低于另外三种模型。综合分析可知,本模型可实现运动行人的抗遮挡,基于监控视频的行人检测跟踪精度明显提高,可实现智能体育视频的双重监测。In view of the problems of small target size and poor detection accuracy and tracking effect,a smart video dual monito-ring algorithm that integrates motion prospect and moving pedestrian is proposed.Firstly,a pedestrian detection model combining with collaborative attention mechanism is constructed based on YOLOv5 algorithm;then using pedestrian detection model as the detector introduces DeepSORT pedestrian tracking algorithm,adding GhostNet-cot re-identification network and anti-occlusion strategy;fi-nally,based on the DeepSORT tracking model,build an intelligent video dual monitoring model with integrated motion prospect and motion pedestrians,so as to realize accurate detection and tracking of moving pedestrians.The experimental results show that the MO-TA,HOTA and IDFI indexes of this model are 48.86%,53.35%and 56.66%respectively,which are higher than the traditional DeepSort model,GAN-RNN model and ECA-BIFPN model.Moreover,the number of IDSW and parameters in this model is 204 and 9.2M,respectively,which is significantly lower than that of the other three models.The comprehensive analysis shows that the model can realize the anti-occlusion of moving pedestrians,and the pedestrian detection and tracking accuracy based on surveillance video is significantly improved,and the dual monitoring of intelligent sports video can be realized.

关 键 词：运动行人检测 YOLOv5算法 DeepSORT 目标跟踪 监控视频 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]