CoMoGCN: Coherent Motion Aware Trajectory Prediction with Graph Representation


Forecasting human trajectories is critical for tasks such as robot crowd navigation and autonomous driving. Modeling social interactions is of great importance for accurate group-wise motion prediction. In this work, we propose a novel framework, coherent motion aware graph convolutional network (CoMoGCN), for trajectory prediction in crowded scenes with group constraints. First, we cluster pedestrian trajectories into groups according to motion coherence. Then, we use graph convolutional networks to aggregate crowd information efficiently. The CoMoGCN also takes advantage of variational inference to capture the variability in the human trajectories by modeling the distribution.

Process flow

The overall process for trajectory prediction includes three steps:

  1. Coherent motion clustering generates labels for each human in an offline data pre-processing procedure.
  2. Intragroup graph and intergroup graph are established for each pedestrian based on the labels.
  3. Trajectory prediction model deploys the intergroup GCN and intragroup GCN to model the intergroup and intragroup interactions.

process flow

Dataset download

Trajectoty dataset with group labels

Sample code

Coherency labels to adjacency matrices (see example of dataset loading in this gist as well):

If you find the dataset or gist useful, please cite the following paper:

title = {CoMoGCN: Coherent Motion Aware Trajectory Prediction with Graph Representation},
author={Chen, Yuying and Liu, Congcong and Shi, Bertram E and Liu, Ming},
booktitle = {BMVC},
year = {2020}