Towards Robust Human Trajectory Prediction in Raw Videos

TL;DR

This paper addresses the challenge of human trajectory prediction in raw videos by proposing a re-tracking method that corrects tracking errors, thereby improving prediction accuracy in real-world scenarios.

Contribution

It introduces a re-tracking algorithm that enhances existing tracking and prediction pipelines by enforcing temporal prediction consistency to mitigate tracking errors.

Findings

Improves tracking accuracy in raw video data.

Enhances trajectory prediction performance.

Effective in challenging real-world scenarios.

Abstract

Human trajectory prediction has received increased attention lately due to its importance in applications such as autonomous vehicles and indoor robots. However, most existing methods make predictions based on human-labeled trajectories and ignore the errors and noises in detection and tracking. In this paper, we study the problem of human trajectory forecasting in raw videos, and show that the prediction accuracy can be severely affected by various types of tracking errors. Accordingly, we propose a simple yet effective strategy to correct the tracking failures by enforcing prediction consistency over time. The proposed "re-tracking" algorithm can be applied to any existing tracking and prediction pipelines. Experiments on public benchmark datasets demonstrate that the proposed method can improve both tracking and prediction performance in challenging real-world scenarios. The code and data are available at https://git.io/retracking-prediction.