基于视频图像驱动的驾驶人注意力估计方法

Method of driver attention estimation based on video image-driven

驾驶人视觉注意力的深入研究对于预测不安全驾驶行为和理解驾驶行为具有重要意义。为此,提出一种基于视频图像驱动的驾驶人注意力估计方法,以估计驾驶人在行车时注意到视域内的行人或车辆等各种对象。该方法利用深度神经网络学习交通场景视频与驾驶员注意力特征之间的映射关系,并融入引导学习模块来提取与驾驶员注意力最相关的特征。考虑到驾驶的动态性,使用动态交通场景视频作为模型输入,设计时空特征提取模块。在稀疏、密集、低照度等常见的交通场景中,将估计的驾驶员注意力模型与收集的驾驶员注意力数据点进行对比。实验结果表明,所提方法能够准确估计驾驶员在驾驶过程中的注意力,对于预测不安全驾驶行为以及促进人们更好地理解驾驶行为具有重要的理论和实用价值。

An in-depth study of drivers' visual attention is important for predicting unsafe driving behavior and understanding driving behavior.A method of driver attention estimation based on video image-driven is proposed to estimate that drivers will notice various objects such as pedestrians or vehicles in the field of view while driving.In the method,the deep neural network is used to learn the mapping relationship between the video of traffic scene and the features of drivers' attention,and the bootstrap learning module is integrated to extract the features that are most relevant to the driver's attention.Considering the dynamicity of driving,a spatio-temporal feature extraction module is designed by using dynamic traffic scene videos as model inputs.The estimated driver attention model is compared with the collected driver attention data points in a variety of common traffic scenes,including sparse,dense,and low-light scenes.The experimental results show that the proposed method can accurately estimate the drivers' attention during driving,and has important theoretical and practical value for predicting unsafe driving behavior and promoting better understanding of driving behavior.