Accuracy of 3-D reconstruction with occlusions

Area of Science:

• Biomechanics
• Motion Analysis
• Computer Vision

Background:

• Three-dimensional (3-D) reconstruction in motion analysis requires multiple camera views.
• Occlusions, where markers are hidden from cameras, introduce errors in 3-D kinematics.
• The impact of occlusions on 2-D images and subsequent 3-D reconstruction needs quantification.

Purpose of the Study:

• To quantify the detrimental effects of occlusions on 2-D images in motion capture.
• To evaluate and recommend signal processing techniques for mitigating occlusion-induced errors.
• To compare the efficacy of 2-D trajectory interpolation versus 3-D trajectory filtering.

Main Methods:

• Collected reference 3-D kinematics data using a three-camera system without occlusions.
• Artificially introduced occlusions into 2-D image data from cyclic motion trials.
• Applied 2-D trajectory interpolation and post-reconstruction filtering (Savitsky-Golay, Butterworth) to correct for occlusions.

Main Results:

• Optimal filter parameters were marker-dependent, indicating variability in correction needs.
• 2-D trajectory interpolation before 3-D reconstruction effectively minimized occlusion-related errors.
• No post-reconstruction filtering was necessary when 2-D interpolation was applied.

Conclusions:

• 2-D trajectory interpolation is a more robust method for handling occlusions than filtering 3-D trajectories.
• Occlusions introduce systematic errors in 3-D reconstructions that are best addressed at the 2-D stage.
• Signal processing recommendations favor pre-reconstruction interpolation for accurate 3-D motion analysis.