Dangerous Driving Behavior Detection with Attention Mechanism
Kun Wang, Xianqiao Chen, Rui Gao · 2019
In order to reduce the incidence of traffic accidents caused by dangerous driving, a dangerous driving behavior recognition model based on convolutional neural network (CNN) and long short-term memory network (LSTM) is proposed. Aiming at the problem of low accuracy of the network model identification, the algorithm is optimized by introducing the unsupervised attention mechanism. The model focuses on a specific visual area and improves the recognition accuracy of the algorithm to some extent by integrating the attention weighted module and the convolution LSTM. The experimental results show that the detection accuracy and detection rate of the algorithm are improved compared with the Two-Stream method and C3D behavior recognition algorithm in the dangerous driving behavior recognition task.