The recognition method of highway vehicle driving dangerous behavior based on CNN-LSTM

To solve the problem of low accuracy in identifying dangerous driving behaviors of highway vehicles, a method for identifying dangerous driving behaviors of highway vehicles based on CNN-LSTM is proposed. Obtaining driving behavior data of highway vehicles through drones equipped with cameras, and extracting keyframe data of dangerous driving behaviors of highway vehicles, Obtain multidimensional features such as spectral characteristics of dangerous driving behavior and temporal spatial characteristics of driving behavior. Using multi-dimensional feature results as input, utilize CNN to enhance feature information fusion, Use the enhanced features as inputs to the LSTM network for temporal analysis, Using the Softmax function, the output of LSTM is converted into probability distributions of different categories of dangerous behaviors to complete the recognition of dangerous driving behaviors on highways. The results indicate that the proposed method has strong ability to identify driving risks of highway vehicles, with MAP index results reaching up to 98% and FPS index results reaching up to 63%, demonstrating high recognition accuracy and efficiency.