Abstract: Predicting the driving behavior of surrounding vehicles is of great significance to improving the safety of driverless cars. This paper divides the driving behavior prediction model in driverless cars into three aspects: description of driving behavior, driving behavior prediction algorithm, and formulation of driving strategy. It sorts out the typical driving behavior prediction models in current driverless vehicles. This article first introduces three driving behavior description methods, subjective definition, finite state machine, and trajectory characteristics, and summarizes the characteristics of various driving behavior prediction models. It points out the limitations of the three typical driving behavior description methods in the actual driving environment. In complex scenes, Hidden Markov model is difficult to classify hidden states and the determination of parameters is too complicated. Finally, the possible application of computer vision in driving behavior prediction models in the future is prospected. This paper summarizes the typical methods of current driving behavior prediction, which provides a reference for data-driven driving behavior prediction research combined with computer vision, and helps researchers to further familiarize themselves with related fields.

Key words: Driverless vehicle; Driving behavior; Prediction model; Driving strategy; Trajectory mapping

摘要： 预测周围车辆的驾驶行为对提高无人驾驶汽车安全性具有重要意义。文章将无人驾驶 汽车中的驾驶行为预测模型分为驾驶行为的描述、驾驶行为预测算法、驾驶策略的制定三个 方面，梳理当前无人驾驶中的典型驾驶行为预测模型。文章首先介绍主观定义、有限状态机、 轨迹特征三种驾驶行为描述方法，并总结了各种驾驶行为预测模型的特点，指出三种典型驾 驶行为描述方法在实际驾驶环境中的局限性，隐马尔科夫模型在复杂场景中存在难以对隐状 态进行分类、参数的确定过于复杂的问题。最后，展望了未来计算机视觉在驾驶行为预测模 型中应用的可能。文章归纳了当前驾驶行为预测的典型方法，为基于数据驱动及与计算机视 觉相结合的驾驶行为预测研究提供参考，有助于研究人员进一步熟悉相关领域。

关键词: 无人驾驶车辆；驾驶行为；预测模型；驾驶策略；轨迹映射