Integrating Datasets for Driving Intention Recognition

Vehicle driving intention recognition is an important component of autonomous driving and advanced driver assistance system (ADAS). By predicting the driving intention of the target vehicle, more information can be provided to human driver or autonomous driving system. Existing vehicle driving intention methods often only consider the lateral driving intentions of vehicles, but recognizing longitudinal driving intentions is also crucial for driving safety. This paper considers the influence of surrounding vehicles on the driving intention of the target vehicle from a microscopic perspective and proposes a model based on self-attention and cross-attention that can simultaneously predict the lateral and longitudinal driving intentions of vehicles. The NGSIM (Next Generation Simulation) dataset and HighD (Highway Drone) dataset are commonly used datasets for vehicle driving intention recognition task, but they both suffer from a significant difference in data quantity for different driving intentions. Therefore, this paper proposes a method to integrate the NGSIM and HighD datasets to balance the number of data for each class of intention. By training our model on this integrated dataset, the model's performance metrics (precision, recall, F1 score) improve by around 4% compared to the baseline model when predicting 0.5s in advance.