Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Half Pound Filter for Real-Time Animation Blending

Riccardo Lasagno

TL;DR

The Half Pound Filter is introduced as a modification of the 1 Euro Filter and algorithms for automatic data-driven tuning and for filter triggering based on motion derivative boundary checks and algorithms for filter triggering based on motion derivative boundary checks are presented.

Abstract

This paper introduces the Half Pound Filter (HPF) as a modification of the 1 Euro Filter (1EF) and algorithms for automatic data-driven tuning and for filter triggering based on motion derivative boundary checks. An application of the filter is presented in the context of human animation replay for real-time simulations, where switches in animation clips cause discontinuities that must be hidden by filtering the bone trajectory without introducing noticeable artifacts. The quality of the filtering will be compared with other common animation filtering techniques using an example case drawn fromthe LaFAN1 dataset, showing that the resulting animation is replayed with higher fidelity by evaluating the Mean Squared Error (MSE) and Normalized Power Spectrum Similarity (NPSS) for each setup. Performances will be evaluated using both a standard predetermined triggerpoint and blending distance and the automatic blending trigger and recovery system.

Half Pound Filter for Real-Time Animation Blending

TL;DR

The Half Pound Filter is introduced as a modification of the 1 Euro Filter and algorithms for automatic data-driven tuning and for filter triggering based on motion derivative boundary checks and algorithms for filter triggering based on motion derivative boundary checks are presented.

Abstract

This paper introduces the Half Pound Filter (HPF) as a modification of the 1 Euro Filter (1EF) and algorithms for automatic data-driven tuning and for filter triggering based on motion derivative boundary checks. An application of the filter is presented in the context of human animation replay for real-time simulations, where switches in animation clips cause discontinuities that must be hidden by filtering the bone trajectory without introducing noticeable artifacts. The quality of the filtering will be compared with other common animation filtering techniques using an example case drawn fromthe LaFAN1 dataset, showing that the resulting animation is replayed with higher fidelity by evaluating the Mean Squared Error (MSE) and Normalized Power Spectrum Similarity (NPSS) for each setup. Performances will be evaluated using both a standard predetermined triggerpoint and blending distance and the automatic blending trigger and recovery system.
Paper Structure (10 sections, 1 equation, 2 figures, 1 table, 2 algorithms)

This paper contains 10 sections, 1 equation, 2 figures, 1 table, 2 algorithms.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Animation Blending in Real-Time Environments
  4. Motion Matching and Data-Driven Techniques
  5. Digital filters for Motion Smoothing
  6. FROM ONE EURO FILTER TO HALF POUND FILTER
  7. TUNING HPF FOR REAL-TIME ANIMATION BLENDING
  8. AUTOMATIC TRIGGER AND RECOVERY POLICY FOR MOTION FILTERS
  9. EVALUATING RESULTS
  10. CONCLUSIONS

Figures (2)

  • Figure 1: The generated example signal and its derivatives. The signal was created by joining a run animation and a fall animation from the LaFAN1 dataset, and the Pitch Local Euler angle for the knee is used. The horizontal dashed lines are the upper and lower boundaries, while the vertical line shows where the two animations have been joined. notice that the only places where the boundaries are violated by acceleration and jerk match the joint frame.
  • Figure 2: Overview of the different filters applied to a noisy signal. The first row shows the resulting smoothing for different techniques applied in a fixed time window, while the second row show the same filters using the automatic trigger and recovery policy.