On Integral Invariants for Effective 3-D Motion Trajectory Matching and Recognition

Area of Science:

• Computer Vision
• Robotics
• Data Analysis

Background:

• Motion trajectories are crucial for analyzing and recognizing human, robot, and object movements.
• Existing methods may struggle with noise and require high-order derivatives.

Purpose of the Study:

• To define new integral invariants for 3D motion trajectories.
• To develop robust and multiscale features for motion analysis and recognition.

Main Methods:

• Designed distance and area integral invariants using kernel functions.
• Estimated area invariants from blurred segments of noisy curves to avoid high-order derivatives.
• Introduced a distance function for trajectory similarity measurement.

Main Results:

• The proposed integral invariants demonstrate computational locality, unique representation, and noise insensitivity.
• Multiscale analysis of motion trajectories is enabled by varying kernel function scales.
• Experiments confirmed the robustness and effectiveness in trajectory matching and sign recognition.

Conclusions:

• The novel integral invariants effectively capture motion cues for trajectory analysis.
• The multiscale approach allows for coarse-to-fine perception of motion features.
• The method shows promise for applications in motion recognition and trajectory similarity tasks.