Learning Scene-Goal-Aware Motion Representation for Trajectory Prediction


Ziyang Ren (Xi'an Jiaotong University), Ping Wei (Xi'an Jiaotong University), Haowen Tang (Xi'an Jiaotong University), Huan Li (Xi'an Jiaotong University), Jin Yang (Xi'an Jiaotong University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

Predicting accurate movement trajectory is a challenging task due to the complexity of human motion patterns and activity scenes. Existing studies focus on extracting motion state information from trajectories but often overlook the representation of future motion trends and interaction with the scene. We present a novel framework called Motion-Scene-Goal Aware Network (MSGANet), which utilizes attention temporal convolutional networks to capture temporal dynamics in motion trajectories. Through self-supervised learning, MSGANet extracts the spatial distribution of motion states. It incorporates multi-scale feature fusion and self-attention mechanisms to extract correlated features between motion states and physical scenes, facilitating inference of goal intentions' spatial distribution. Additionally, MSGANet employs cross-attention mechanisms to enable feature interactions between motion states and goal intentions. By integrating scene semantic aware fusion and aware interaction of goal intentions, it enhances the representation of motion state features for predicting future motion trends. Experiments on ETH-UCY and SDD datasets prove the strength of our method.

Citation

@inproceedings{Ren_2024_BMVC,
author    = {Ziyang Ren and Ping Wei and Haowen Tang and Huan Li and Jin Yang},
title     = {Learning Scene-Goal-Aware Motion Representation for Trajectory Prediction},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0228.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection