Neural Temporal Models for Human Motion Prediction
Open Access
- Author:
- Gopalakrishnan, Anand
- Graduate Program:
- Electrical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- March 28, 2019
- Committee Members:
- David Jonathan Miller, Thesis Advisor/Co-Advisor
Clyde Lee Giles, Thesis Advisor/Co-Advisor
Minghui Zhu, Committee Member - Keywords:
- Motion Prediction
Recurrent Neural Networks
Deep Learning
Motion Synthesis
Human Motion Evaluation - Abstract:
- This work proposes novel neural temporal models for predicting and synthesizing human motion, achieving state-of-the-art in modeling long-term motion trajectories while being competitive with prior work in short-term prediction, with significantly less computational expense. Key aspects of the proposed system include: 1) a novel, two-level processing architecture that helps in generating "guiding" trajectories, 2) a set of easily computable features that incorporate motion derivative information into the model, and 3) a novel multi-objective loss function that helps the model to incrementally progress from the simpler task of next-step prediction to the harder task of multi-step closed-loop prediction. The results demonstrate that these innovations facilitate improved modeling of long-term motion trajectories. Finally, a novel metric, called Normalized Power Spectrum Similarity (NPSS) is proposed, to evaluate the long-term predictive ability of motion synthesis models, complementing the popular mean-squared error (MSE) measure of the Euler joint angles over time. A user study is conducted to determine if the proposed NPSS correlates with human evaluation of long-term motion more strongly than MSE and finds that it indeed does.