基于信息瓶颈理论的驾驶员分心行为识别

Driver Distracted Behavior Recognition Based on Information Bottleneck Theory

摘要:

针对驾驶员分心行为识别问题，将信息瓶颈理论与图卷积网络相结合，提出一个基于二维姿态估计的动作识别网络，增加神经网络对有效信息的保留程度，从而弥补输入信息量的不足。基于通道级拓扑细化图卷积网络，在有限输入信息下实现了准确的动作识别。

Abstract:

Aiming at the problem of driver distracted behavior recognition， the information bottleneck theory and the graph convolutional network were combined to realize the action recognition based on the 2D pose estimation， which effectively increases the retention degree of neural network for effective information， so as to make up for the lack of input information. The accurate action recognition was achieved with the limited input information in combination with CTR-GCN.