Existing data-driven lane change trajectory prediction methods lack the capability to achieve broad consistency in the complex and variable real-world traffic scenarios due to their reliance on training with pre-collected static datasets from fixed scenarios [13]. In reality, the scenarios vary widely, and there is no assurance of sufficient training data for each scenario. The primary challenge is to effectively apply knowledge from observed scenarios to unknown ones while ensuring the method adapts quickly to limited data without compromising prediction accuracy. Based on the aforementioned research gap, this paper develops a novel data-driven lane change prediction method which enhances the transfer of knowledge from observed to unknown scenarios using the few-shot learning (FSL) concept. Our proposed method enables a pre-trained LSTM model to be quickly deployed in new scenarios with only a few samples. Extensive experiments have been conducted using the NGSIM dataset to demonstrate the performance of our method.
Transferring Knowledge from Observed to Unknown: A Data-Driven Lane Change Trajectory Prediction Strategy Based on the Concept of Few-Shot Learning
2024-09-24
1033009 byte
Conference paper
Electronic Resource
English
Few-Shot traffic prediction based on transferring prior knowledge from local network
Taylor & Francis Verlag | 2023
|LANE CHANGE TRAJECTORY PREDICTION BY USING RECORDED HUMAN DRIVING DATA
British Library Conference Proceedings | 2013
|