A Siamese Neural Network-Based Method to Recognize the Abnormal Driving Behaviors for Autonomous Vehicles

It is such a challenge to ensure a safety driving when the self-driving vehicles are in a mixed traffic environment with human-driven vehicles. There are some abnormal driving behaviors, which could lead to accidents, financial losses, and even casualties. Therefore, the identification of abnormal driving behaviors is vital for the safety of autonomous vehicles, which could help to avoid the potential dangers. Considering that the abnormal behaviors take a small part of driving data, this paper proposed a Siamese neural network recognition model, which could be built up with small datasets. In detail, the vehicle trajectory data were preprocessed and used to train and validate the proposed method. Experimental results indicated that the Siamese neural network-based method could effectively identify abnormal driving behaviors with an accuracy rate of 88.0%. To some extent, the proposed method could assist the autonomous vehicles to recognize the abnormal driving behaviors successfully with a small sample size of driving data.