Plausible Vehicle Trajectory Augmentation With Hybrid Machine Learning

In recent decades, advancements in autonomous driving technology have fundamentally transformed the automotive industry and transportation systems. Autonomous vehicles hold the potential to significantly reduce traffic accidents, improve driving convenience, and alleviate congestion. A crucial factor in achieving these outcomes is the accurate prediction of surrounding vehicle trajectories, which relies on the effective application of artificial intelligence and machine learning algorithms. However, the collection of sufficient high-quality driving data for training these models poses practical challenges, particularly in complex and unpredictable traffic scenarios. To address this issue, this paper proposes a novel data augmentation technique, integrating Chebyshev polynomial fitting with Hierarchical Density-Based Spatial Clustering of Applications with Noise clustering and Gaussian mixture models to generate robust training data. This study focuses on augmenting vehicle driving datasets by automatically managing variations in time-series lengths and the clustering of driving maneuvers. This approach ultimately generates augmented data with a high degree of similarity to real-world data. Validation on the inD dataset confirms that our approach excels in expanding training datasets, yielding substantial improvements in trajectory prediction accuracy for autonomous vehicles when applied to a vanilla long short-term memory model. This study outlines the methodology and validates the effectiveness of the augmented data, offering a promising method for enhancing autonomous vehicle training frameworks.