Advancing Spatiotemporal Learning for Cyclist Safety and Collision Forecasting in Autonomous Driving
Loading...
Authors
Desai, Nishq Poorav
Date
2025-02-28
Type
thesis
Language
eng
Keyword
Cyclist safety , Collision forecasting , Dataset , Transformer
Alternative Title
Abstract
In this work, we address the under-representation of cyclist data and limitations in TTC prediction. We present CycleCrash, a novel dataset consisting of 3,000 dashcam videos with 436,347 frames that capture cyclists in various critical situations, from collisions to safe interactions. This dataset enables 9 different tasks focusing on potentially hazardous conditions for cyclists and is annotated with collision-related, cyclist-related, and scene-related labels. Next, we present VidNeXt, a novel method that utilizes ConvNeXt to learn spatial components and a non-stationary transformer to encode the temporal aspects of videos for the defined tasks within CycleCrash. We apply VidNeXt on CycleCrash and compare with 7 baselines and a detailed ablation.
Moreover, TTC is a challenging task, requiring the models to extract features from short and long-range dependencies in spatial and temporal dimensions. While SOTA methods are not designed to focus on this, we propose CollideNet, a novel two-stream method that incorporates a multi-scale hierarchical architecture and disentangles trend, seasonality, and non-stationary components for more effective feature extraction. We compare CollideNet with 12 baselines and conduct a detailed ablation study on 3 benchmark datasets. Collectively, these contributions tackle key challenges in cyclist safety prediction and TTC forecasting by introducing a novel dataset and methods that enhance model accuracy and generalization.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 4.0 International
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 4.0 International