The article presents a framework, ARCANE, for generating synthetic datasets to improve models predicting pedestrian intentions, crucial for autonomous driving. The lack of diverse datasets with crossing and non-crossing scenarios contributes to the challenge of creating accurate and fast models. ARCANE addresses this by generating a large, varied dataset named PedSynth. The researchers also propose a deep model, PedGNN, with a low memory footprint for onboard deployment. PedGNN uses a GNN-GRU architecture that takes a sequence of pedestrian skeletons as input to predict crossing intentions. The ARCANE framework, PedSynth dataset, and PedGNN model will be publicly released.
Publication date: 15 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.06757
