基于级联卷积神经网络的驾驶员分心驾驶行为检测

Driver Distracted Driving Behavior Detection Based on Cascaded Convolutional Neural Network

针对驾驶员分心驾驶行为检测,设计一种级联卷积神经网络检测框架。检测框架由第一级分心行为预筛选卷积网络和第二级分心行为精确检测卷积网络两个全卷积网络级联构成。预筛选卷积网络是一个轻量级的图像分类网络,负责对原始数据进行快速筛选,其网络层数少、训练速度快,结构特征冗余较少,能够减少后续网络的计算负担;分心行为精确检测卷积网络采用VGG(Visual geometry group)模型特征提取的深度迁移学习检测算法网络,通过迁移学习重新训练分类器和部分卷积层。提出的级联神经网络最终可以实现9种驾驶员分心驾驶行为的准确识别检测。实验结果表明,相比主流单模型检测方法,在保证算法效率的同时准确率均有明显提升,准确率达到93.3%,有效降低了误检率。该方法具有较好的鲁棒性和泛化能力。

A cascaded convolutional neural network(CCNN)detection framework was designed for driver distracted driving behavior detection.The detection framework consisted of a first-stage distraction behavior pre-screening convolution network and a second-stage distraction behavior accurate detection convolution network.Pre-filtered convolution network was a lightweight image classification network,which had responsible for fast filtering the original data.The network layer number was small,the training speed was fast,and the redundancy of the structural features was small,so it can reduce the computational burden of the subsequent network.The CCNN adopted the deep migration learning detection algorithm network of VGG(visual geometry group)model feature extraction,and retrained classifiers and partial convolution layers through migration learning.Finally,the cascaded neural network proposed can realize the accurate identification and detection of nine kinds of distracted driving behaviors.The experimental results show that compared with the mainstream single model detection method,the accuracy of the algorithm is improved obviously,the accuracy is 93.3%,and the false detection rate is effectively reduced.The method has better robustness and generalization ability.