摘要
在遮挡物较多的变电站场景下,传统的目标跟踪算法容易出现人员跟丢、身份变换、目标被遮挡无法识别等情况,无法做到对目标的准确实时跟踪。针对此问题,提出了融合度量学习与卡尔曼滤波的变电站运动目标跟踪方法。首先,采用融合多尺度特征的YOLOv3算法检测目标,利用带权值的匈牙利算法匹配目标的历史运动轨迹;然后,通过卡尔曼滤波进行轨迹预测,并应用度量学习匹配预测轨迹与历史运动轨迹,结合目标外观特征信息,以实现变电站内运动目标的跟踪;最后,以变电站场景下的人员为例进行实验,结果表明,人员跟踪准确率高,且能满足变电站应用场景下的鲁棒性和实时性要求。
The traditional target tracking algorithm is easy to have tracking loss,identity transformation and other situations in the substation scene with more obstructions,so it can't track the target accurately in real time.To solve this problem,a method of moving target tracking in substation is proposed,which combines metric learning and Kalman filter.In this method,firstly,You Only Look Once algorithm is used to detect the target,and then Hungary algorithm with weights is used to match the historical trajectory of the target,then Kalman filter is used to predict the trajectory,and metric learning is used to match the predicted trajectory with the historical trajectory,and the tracking of the moving target is realized in the substation combined with the appearance feature information of the target.Taking the personnel in the substation as an example,the experiment shows that the method has high tracking accuracy and can meet the requirements of robustness and real-time in the substation application scenario.
作者
雷景生
杨忠光
LEI Jingsheng;YANG Zhongguang(School of Computer Science and Technology,Shanghai University of Electric Power,Shanghai200082,China)
出处
《上海电力大学学报》
CAS
2022年第1期40-47,共8页
Journal of Shanghai University of Electric Power
基金
国家自然科学基金(61672337)。
