Abstract:With the fast development of AI-related techniques, the applications of trajectory prediction are no longer limited to easier scenes and trajectories. More and more heterogeneous trajectories with different representation forms, such as 2D or 3D coordinates, 2D or 3D bounding boxes, and even high-dimensional human skeletons, need to be analyzed and forecasted. Among these heterogeneous trajectories, interactions between different elements within a frame of trajectory, which we call the ``Dimension-Wise Interactions'', would be more complex and challenging. However, most previous approaches focus mainly on a specific form of trajectories, which means these methods could not be used to forecast heterogeneous trajectories, not to mention the dimension-wise interaction. Besides, previous methods mostly treat trajectory prediction as a normal time sequence generation task, indicating that these methods may require more work to directly analyze agents' behaviors and social interactions at different temporal scales. In this paper, we bring a new ``view'' for trajectory prediction to model and forecast trajectories hierarchically according to different frequency portions from the spectral domain to learn to forecast trajectories by considering their frequency responses. Moreover, we try to expand the current trajectory prediction task by introducing the dimension $M$ from ``another view'', thus extending its application scenarios to heterogeneous trajectories vertically. Finally, we adopt the bilinear structure to fuse two factors, including the frequency response and the dimension-wise interaction, to forecast heterogeneous trajectories via spectrums hierarchically in a generic way. Experiments show that the proposed model outperforms most state-of-the-art methods on ETH-UCY, Stanford Drone Dataset and nuScenes with heterogeneous trajectories, including 2D coordinates, 2D and 3D bounding boxes.

Synthesizing Human Trajectories Based on Variational Point Processes

Variational Sampling of Temporal Trajectories

GeoGail: A Model-based Imitation Learning Framework for Human Trajectory Synthesizing

Context-Aware Variational Trajectory Encoding And Human Mobility Inference

Conditional Variational Inference for Multi-modal Trajectory Prediction with Latent Diffusion Prior.

A Deep Generative Model for Trajectory Modeling and Utilization.

Enhanced Multimodal Trajectory Prediction for Autonomous Vehicles Using Advanced Diffusion Model Techniques

A Spatial-Temporal Topic Segmentation Model For Human Mobile Behavior

An Efficient Spatial-Temporal Model Based on Gated Linear Units for Trajectory Prediction

Uncovering the Missing Pattern: Unified Framework Towards Trajectory Imputation and Prediction

Two-Stage Trajectory Generation Model for Realistic Human Mobility Simulation.

A Variational Point Process Model for Social Event Sequences

CMTS: Conditional Multiple Trajectory Synthesizer for Generating Safety-critical Driving Scenarios

A Novel Trajectory Generator Based on a Constrained GAN and a Latent Variables Predictor

Revisiting Synthetic Human Trajectories: Imitative Generation and Benchmarks Beyond Datasaurus

Another Vertical View: A Hierarchical Network for Heterogeneous Trajectory Prediction via Spectrums

TrajVAE: A Variational AutoEncoder model for trajectory generation

Modeling and classifying human activities from trajectories using a class of space-varying parametric motion fields

Trajectory-User Linking Via Variational AutoEncoder.

A Unified 3D Human Motion Synthesis Model Via Conditional Variational Auto-Encoder

TrajGPT: Controlled Synthetic Trajectory Generation Using a Multitask Transformer-Based Spatiotemporal Model