摘要
针对基于序列建模的车辆异常轨迹检测方法轨迹空间特征提取不够充分而降低了检测效果这一问题,提出融合时序和空间特征的车辆异常轨迹检测方法,充分提取轨迹的时间与空间特征以提升异常轨迹检测精度。采用融合自注意力机制的堆叠序列自编码器,从网格化后的映射轨迹中提取轨迹时序特征;引入全连接神经网络,提取轨迹偏转量和行驶距离等空间特征;融合轨迹的时间和空间特征,进行异常轨迹检测以提升检测效果。实验表明,提出的方法在真实出租车数据集上的异常轨迹检测准确率优于92%,F1评分优于80%,与XGBoost、IBAT、ATDC和ATD-RNN方法相比,检测性能提升较为明显。
The vehicle abnormal trajectory detection method based on sequence modeling can effectively extract temporal features,but the spatial features are not sufficiently extracted,which reduces the detection effect to some extent.To address this problem,in order to fully extract temporal and spatial features of trajectories to improve the accuracy of abnormal trajectory detection,this paper proposes a vehicle abnormal trajectory detection method that fuses temporal and spatial features.A stacked sequence auto-encoder with a fused self-attention mechanism is used to extract the trajectory timing features from the mapped trajectory after gridding,and a fully connected neural network is introduced to extract the spatial features such as trajectory deflection and distance,and fuse the temporal and spatial features of the trajectory for abnormal trajectory detection to improve the detection effect.Experiments show that the detection accuracy of the method is higher than 92%and the F1 score is higher than 80%on the real cab dataset,and the detection performance is significantly improved compared with XGBoost,IBAT,ATDC and ATD-RNN methods.
作者
夏英
张安洁
XIA Ying;ZHANG Anjie(School of Computer Science and Technology,The Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2023年第2期202-209,共8页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(41971365)
重庆市高技术产业重大产业技术研发项目(D2018-82)
重庆市教委重点合作项目(HZ2021008)。
关键词
异常轨迹检测
序列自编码器
自注意力机制
特征融合
abnormal trajectory detection
sequence auto-encoder
self-attention mechanism
feature fusion
