The invention discloses a self-supervision vehicle abnormal track detection method based on vehicle-road cooperation. The method comprises the following steps: 1) modeling road information into a graph structure; each road section in the network represents a node, and connectivity between the road sections is represented by edges in the graph. 2) selecting a sub-graph view for a target node, and selecting a sub-graph with a high intimacy score with the target node from sub-graphs generated by random walk as input; 3) generating a low-dimensional vector representation of a target node and a corresponding sub-graph; 4) obtaining a generation anomaly score of the node through a generative self-supervised learning task, and reflecting the generation anomaly score as a regression loss between a reconstruction feature vector generated by a decoder and an original feature vector; 5) obtaining a comparison abnormal score of the node through a comparative self-supervised learning task, and respectively scoring by generating positive and negative examples of a target node by utilizing the structural information of the input graph to obtain a comparison loss of the node; 6) calculating a final abnormal score according to the generated abnormal score and the compared abnormal score, and 7) setting a threshold value according to the abnormal score, and judging the node exceeding the threshold value to be abnormal. The method can find out the node with the abnormal behavior under the condition of not using a label.
本发明公开一种基于车路协同的自监督车辆异常轨迹检测方法,1)将道路信息建模成图结构;网络中的每个路段代表一个节点,路段之间的连通性由图中的边表示。2)为目标节点选取子图视图,从随机游走生成的子图中选取与目标节点亲密度分数高的子图作为输入;3)生成目标节点和对应子图的低维向量表示;4)通过生成性自监督学习任务得到节点的生成异常分数,反映为由解码器生成得到的重建特征向量与原始特征向量之间的回归损失;5)通过对比性自监督学习任务得到节点的对比异常分数,利用输入图的结构信息通过生成目标节点的正负例分别打分得到节点的对比损失;6)由生成异常分数和对比异常分数计算得到最终异常得分;7)根据异常分数设定一个阈值,将超过该阈值的节点判定为异常。本发明可在不使用标签的情况下找出异常行为的节点。
Self-supervision vehicle abnormal track detection method based on vehicle infrastructure cooperation
一种基于车路协同的自监督车辆异常轨迹检测方法
2024-09-20
Patent
Electronic Resource
Chinese
IPC: | G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung |
European Patent Office | 2024
|Vehicle auxiliary driving method based on vehicle-infrastructure cooperation
European Patent Office | 2023
|European Patent Office | 2021
|ABNORMAL VEHICLE DETECTION SERVER AND ABNORMAL VEHICLE DETECTION METHOD
European Patent Office | 2022
|