The invention relates to a probability calculation method model based on deep learning. A traffic accident feature vector is constructed by summarizing basic data such as a time period, a vehicle speed and a road section direction, and two probability calculation method states for determining whether a traffic accident occurs are determined at the same time. In a training stage, for a noise-free self-encoding network method in deep learning, hidden layer parameters capable of representing deep features are obtained from a label-free data set, and a new feature set is generated. In a probability calculation method stage, for a new data set with a label, Softmax regression learning is utilized to generate a probability calculation method classifier of a traffic accident. For the setting of a learning rate, the performance of the probability calculation method in the traffic accident probability calculation method is compared between the manual setting and the learning rate attenuation strategy based on exponential attenuation through experiments. Experiments show that when the learning rate attenuation strategy based on exponential attenuation is applied to the traffic accident probability calculation method, relatively good stability is achieved, and meanwhile the accuracy rate ranges from 88% to 89%.

    本发明涉及一种基于深度学习的概率计算方法模型。通过总结出时间段、车速及路段方向等基础数据构建交通事故特征向量,同时确定了是否发生交通事故的两种概率计算方法状态。在训练阶段,对于深度学习中的无噪自编码网络方法,从无标签数据集中获得可表征深层特征的的隐藏层参数并生成新的特征集合。在概率计算方法阶段,对于有标签的新数据集,我们将利用Softmax回归学习生成交通事故的概率计算方法分类器。对于学习率的设定,通过实验比较人工设定与基于指数衰减的学习率衰减策略在交通事故概率计算方法中的概率计算方法性能。实验发现,将基于指数衰减的学习率衰减策略应用于交通事故概率计算方法具有相对较好的稳定性,同时准确率为88%到89%之间。


    Access

    Download


    Export, share and cite



    Title :

    Method for calculating probability of occurrence of traffic accident based on deep learning


    Additional title:

    一种基于深度学习的计算交通事故发生概率的方法


    Contributors:
    ZHOU WEIJIAN (author) / WANG SHIKUN (author) / SUN TONGXIN (author)

    Publication date :

    2021-06-29


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06K Erkennen von Daten , RECOGNITION OF DATA




    Towards Deep Learning based Traffic Accident Analysis

    Naseer, Atif / Nour, Mohamed K. / Alkazemi, Basem Y. | IEEE | 2020


    Modeling traffic accident occurrence and involvement

    Abdel-Aty, Mohamed A. | Online Contents | 2000


    Multisource Accident Datasets-Driven Deep Learning-Based Traffic Accident Portrait for Accident Reasoning

    Chun-Hao Wang / Yue-Tian-Si Ji / Li Ruan et al. | DOAJ | 2024

    Free access

    Traffic accident occurrence area prediction method and device

    WU RAN / WANG LIN / BAO XIANGWEN et al. | European Patent Office | 2024

    Free access