TrajPRed: Trajectory Prediction With Region-Based Relation Learning (2024-04-10T00:00:00.000000Z)

TL;DR

A region-based relation learning paradigm that models social interactions via region-wise dynamics of joint states, i.e., the changes in the density of crowds, and it is shown that diverse prediction benefits from region-based relation learning.

Abstract

Forecasting human trajectories in traffic scenes is critical for safety within mixed or fully autonomous systems. Human future trajectories are driven by two major stimuli, social interactions, and stochastic goals. Thus, reliable forecasting needs to capture these two stimuli. Edge-based relation modeling represents social interactions using pairwise correlations from precise individual states. Nevertheless, edge-based relations can be vulnerable under perturbations. To alleviate these issues, we propose a region-based relation learning paradigm that models social interactions via region-wise dynamics of joint states, i.e., the changes in the density of crowds. In particular, region-wise agent joint information is encoded within convolutional feature grids. Social relations are modeled by relating the temporal changes of local joint information from a global perspective. We show that region-based relations are less susceptible to perturbations. In order to account for the stochastic individual goals, we exploit a conditional variational autoencoder to realize multi-goal estimation and diverse future prediction. Specifically, we perform variational inference via the latent distribution, which is conditioned on the correlation between input states and associated target goals. Sampling from the latent distribution enables the framework to reliably capture the stochastic behavior in test data. We integrate multi-goal estimation and region-based relation learning to model the two stimuli, social interactions, and stochastic goals, in a prediction framework. We evaluate our framework on the ETH-UCY dataset and Stanford Drone Dataset (SDD). We show that diverse prediction benefits from region-based relation learning. The predicted intermediate location distributions better fit the ground truth when incorporating the relation module. Our framework outperforms the state-of-the-art models on SDD by 27.61%/18.20% of ADE/FDE metrics. Our code is available at https://github.com/olivesgatech/TrajPRed

References67 items

Trajectory Forecasting Based on Prior-Aware Directed Graph Convolutional Neural Network

Patient Aware Active Learning for Fine-Grained OCT Classification

Volumetric Supervised Contrastive Learning for Seismic Semantic Segmentation

Multi-Modal Learning Using Physicians Diagnostics for Optical Coherence Tomography Classification

A Gating Model for Bias Calibration in Generalized Zero-Shot Learning

Explanatory Paradigms in Neural Networks: Towards relevant and contextual explanations

CSCNet: Contextual Semantic Consistency Network for Trajectory Prediction in Crowded Spaces

Open-Set Recognition With Gradient-Based Representations

A Survey on Human-aware Robot Navigation

Joint learning for spatial context-based seismic inversion of multiple data sets for improved generalizability and robustness

Spatio-Temporal Graph Dual-Attention Network for Multi-Agent Prediction and Tracking

On the Structures of Representation for the Robustness of Semantic Segmentation to Input Corruption

AutoTrajectory: Label-free Trajectory Extraction and Prediction from Videos using Dynamic Points

Imitative Non-Autoregressive Modeling for Trajectory Forecasting and Imputation

It Is Not the Journey but the Destination: Endpoint Conditioned Trajectory Prediction

Training

Trajectron++: Dynamically-Feasible Trajectory Forecasting with Heterogeneous Data

The Garden of Forking Paths: Towards Multi-Future Trajectory Prediction

An Exponential Learning Rate Schedule for Deep Learning

Analyzing the Variety Loss in the Context of Probabilistic Trajectory Prediction

Human motion trajectory prediction: a survey

Conditional Generative Neural System for Probabilistic Trajectory Prediction

Dynamic Channel: A Planning Framework for Crowd Navigation

Peeking Into the Future: Predicting Future Person Activities and Locations in Videos

The Trajectron: Probabilistic Multi-Agent Trajectory Modeling With Dynamic Spatiotemporal Graphs

Scene-LSTM: A Model for Human Trajectory Prediction

SoPhie: An Attentive GAN for Predicting Paths Compliant to Social and Physical Constraints

Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks

Social Attention: Modeling Attention in Human Crowds

What will Happen Next? Forecasting Player Moves in Sports Videos

Context-Aware Trajectory Prediction

Learning Social Etiquette: Human Trajectory Understanding In Crowded Scenes

Social LSTM: Human Trajectory Prediction in Crowded Spaces

Pedestrian's Trajectory Forecast in Public Traffic with Artificial Neural Networks

Socially-Aware Large-Scale Crowd Forecasting

Will the Pedestrian Cross? A Study on Pedestrian Path Prediction

Who are you with and where are you going?

You'll never walk alone: Modeling social behavior for multi-target tracking

Abnormal crowd behavior detection using social force model

Crowds by Example

Social force model for pedestrian dynamics.

Plans and Situated Actions: The Problem of Human Machine Communication.

AUTO-ENCODING VARIATIONAL BAYES

IEEE Transactions on Intelligent Transportation Systems

GENERATIVE ADVERSARIAL NETS

and the M.Sc. degree in artificial intelligence from Bu-Ali Sina University and the Ph.D. degree in computer science from UTA

His dissertation focused on single-shot face recognition using deep learning algorithms for security applications

Forgetting: A Novel Approach to Explain and Interpret Deep Neural Networks in Seismic Interpretation

Signal Processing: Image Communications from an Area Editor for Columns and Forums in IEEE from 2009 to 2012

Georgia Tech in 2023

an LSTM-based encoder-decoder model. The latent features of spatially proximal agents are pooled to model social interactions

in contrast to deterministic methods, S-GAN introduces a generative adversarial learning paradigm based on S-LSTM to generate plausible trajectory outcomes

based on LSTM-GAN, the model employs both scene and social attention mechanisms to account for the scene as well as social compliance

received the B.E. degree in Software

medical imaging, and subsurface imaging

uncertainty and trust, and human-in-the-loop algorithms

trajectory, the future positions are generated by on this optimal goal

a context-aware transfer model that alleviates the discrepancies of physical and social interactions across scenes

minimum

a vanilla LSTM model leveraging only the agent history information

Linear : a linear regression model that estimates parameters by minimizing the least square error

of several journal publications

a stochastic goal-conditioned method that incorporates social influence from neighboring agents

S-LSTM

closest to the ground truth with minimum ADE for evaluation

programmatic labeling solutions for computer vision tasks, continuing to drive innovation in machine learning, and artificial intelligence

his tenure at Ford, where he held the position of Director of Perception, leading the development of perception algorithms for L2+ autonomy