Abstract: Accurate recognition of driving behavior plays an important role in vehicle driving safety. Aiming at the subjectivity and low recognition accuracy of current driving behavior recognition methods, this paper establishes a driving behavior recognition model based on the gated recurrent unit. The driving simulator is used to obtain driver manipulation data and establish a driving behavior data set. After the model is trained, it can effectively use the timing characteristics of the driver manipulation data to successfully recognize driving behavior. The recognition accuracy of the model reaches 95.24%, which provides important theoretical support for the field of traffic safety.

Key words: Driving behavior recognition; Deep learning; Gated recurrent unit

摘要： 准确识别驾驶行为，有利于提高车辆行驶安全性。文章针对目前驾驶行为识别方法的主观 性和识别准确率低的问题，从驾驶员的角度出发，建立了基于门控循环单元网络的驾驶行为辨识 模型。利用驾驶模拟器获取驾驶员操纵数据并建立驾驶行为数据集，模型经过训练后，能够有效 利用驾驶员操纵数据的时序特征，成功识别紧急加速、紧急制动和平稳直线行驶，模型的识别准 确率到达 96.67%，为交通安全领域提供重要的理论支持。

关键词: 驾驶行为辨识；深度学习；门控循环单元