Intelligent Driving Behavior Recognition and Legal Liability Issues Using Deep Learning Convolutional Networks

Abstract

This work aimed to develop a driving behavior recognition and liability assistance determination method applicable to practical traffic safety management and criminal liability determination scenarios. First, an improved deep neural network was designed, which integrated multi-scale 3D convolutional structures and attention mechanisms to efficiently extract driving behavior features from both spatial and temporal dimensions. Next, a subset of six typical driving behaviors was constructed based on the Drive&Act public dataset, followed by sample labeling and feature preprocessing. Finally, based on behavior recognition outputs, a legal article logic mapping model and a behavior risk scoring mechanism were proposed to quantify the legal liability risks corresponding to driving behaviors.