数据驱动下交通异常行为的双层识别模型

为实现对车辆异常行为的准确识别，提高车辆行驶的安全性，提出了一种基于支持向量机(support vector machine, SVM)、监督学习与长短期记忆(long short term memory, LSTM)深度学习的交通异常驾驶行为双层识别模型。首先，对车辆轨迹数据筛除和滤波，构建异常行为数据集；其次，从异常行为轨迹特征中提取出特定异常行为的特征标签，并人为标定在训练集中；再次，构建SVM模型对训练集进行粗识别，基于SVM的二分法原理，从测试集中筛选出异常行为；最后，通过LSTM时间序列模型构建具体种类的异常行为模型，并通过深度学习的方法，从异常行为数据中细分为蛇形驾驶、急速变向、侧滑、大半径转弯、快速U型转弯、急刹车等具体的异常驾驶行为。本次实验选用下一代仿真(next generation simulation, NGSIM)数据中US-101高速公路和peachtree城市道路的数据集的轨迹数据验证SVM和LSTM双层识别模型的性能，包括均方根误差、识别准确率等。结果表明，构建的双层识别模型在第一层有98%的识别准确率，第二层有超过80%的识别准确率，可以较为准确地识...

Double Layer Identification Model of Traffic Abnormal Behavior Driven by Data

To realize the accurate identification of vehicle abnormal behavior and improve the safety of vehicle driving, an abnormal driving behavior recognition model based on SVM supervised learning and LSTM deep learning distribution was proposed. Firstly, the vehicle trajectory data were filtered to construct the abnormal behavior data set. Secondly, based on the trajectory characteristics of abnormal behaviors, the feature labels of specific abnormal behaviors were calibrated in the training set extracted and artificially. Thirdly, several SVM models were constructed to roughly identify the training set, and abnormal behaviors were screened from the test set based on the dichotomy principle of SVM. Finally, the LSTM model was used to construct specific types of abnormal behavior models, and through the method of deep learning, the abnormal behavior data were subdivided into specific abnormal driving behaviors such as weaving, swerving, sideslippling, turing with a wide radius, fast U-turn, and sudden-braking. The trajectory data of US-101 highway and peachtree urban road in NGSIM was applied to verify the performance of SVM and LSTM double recognition models, using root mean square error and recognition accuracy, etc. The results show that the constructed double-layer recognition model has 98 % recognition accuracy in the first layer and more than 80 % recognition accuracy in the second layer, this model can accurately identify abnormal driving behaviors in big data and determine their types.