To achieve accurate recognition of surrounding vehicles' driving intentions for autonomous driving vehicles, this paper proposes a driving intent recognition model based on future trajectory feature extraction and Extreme Gradient Boosting algorithm (XGBoost), termed as Bi-LSTM-XGBoost. Firstly, using real road datasets from NGSIM (Next Generation SIMulation) US101 and I–80 segments, the driving intentions of vehicle historical trajectories are annotated. Then, the future trajectories of target vehicles are predicted using the Bi-directional Long Short-Term Memory (Bi-LSTM) module, and during the training process, a certain proportion of labeled data is used as model input for feature extraction to prevent the influence of incorrect outputs on subsequent results. Secondly, a driving intent recognition framework is constructed by unwinding multidimensional features obtained in the previous step to one dimension, and XGBoost module is utilized to integrate historical and future trajectory outputs for driving intent recognition. Finally, experimental results demonstrate that this method achieves an accuracy of 97.4% in recognizing driving intentions at the prediction horizon of 3 seconds for a historical trajectory of 5 seconds, indicating its strong capability in driving intent recognition.
Driving Intention Recognition Based on Bi-LSTM Network and XGBoost
09.05.2025
884086 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Vehicle lane changing intention recognition method based on TCN-LSTM network
Europäisches Patentamt | 2024
|Driving intention reasoning method for LSTM network optimization based on complete transfer learning
Europäisches Patentamt | 2023
|