Multiclass-SGCN: Sparse Graph-based Trajectory Prediction with Agent Class Embedding

TL;DR

Multiclass-SGCN introduces a sparse graph convolution approach that incorporates agent types and adaptive interaction masking to improve multi-class trajectory prediction accuracy in complex road scenarios.

Contribution

It presents a novel sparse graph convolution network with an interaction mask for multi-class trajectory prediction considering agent types and interactions.

Findings

Outperforms state-of-the-art methods on Stanford Drone Dataset.

Provides more realistic and plausible trajectory predictions.

Effectively models multi-class interactions with a sparse graph approach.

Abstract

Trajectory prediction of road users in real-world scenarios is challenging because their movement patterns are stochastic and complex. Previous pedestrian-oriented works have been successful in modelling the complex interactions among pedestrians, but fail in predicting trajectories when other types of road users are involved (e.g., cars, cyclists, etc.), because they ignore user types. Although a few recent works construct densely connected graphs with user label information, they suffer from superfluous spatial interactions and temporal dependencies. To address these issues, we propose Multiclass-SGCN, a sparse graph convolution network based approach for multi-class trajectory prediction that takes into consideration velocity and agent label information and uses a novel interaction mask to adaptively decide the spatial and temporal connections of agents based on their interaction scores. The proposed approach significantly outperformed state-of-the-art approaches on the Stanford Drone Dataset, providing more realistic and plausible trajectory predictions.