Deform PF-MT: particle filter with mode tracker for tracking nonaffine contour deformations

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Tracking deforming object contours in image sequences is crucial for various applications.
• Challenges include large non-affine deformations, clutter, occlusions, low contrast, and outlier imagery.
• Traditional particle filters (PF) are computationally expensive for high-dimensional contour deformation spaces.

Purpose of the Study:

• To develop an efficient algorithm for tracking deforming object contours under challenging conditions.
• To address the limitations of existing particle filtering methods in high-dimensional state spaces.
• To adapt the particle filtering with mode tracking (PF-MT) approach for contour deformation tracking.

Main Methods:

• Proposed a novel algorithm, deform PF-MT, building upon particle filtering with mode tracking.
• Utilized the low-dimensional effective basis space of deformation at subsampled locations.
• Adapted PF-MT for the non-Euclidean, infinite-dimensional space of contours, requiring significant modifications.

Main Results:

• The deform PF-MT algorithm demonstrates effective contour tracking even with large deformations and image imperfections.
• The approach efficiently manages the high dimensionality of contour deformation spaces.
• Successfully adapted PF-MT for the complexities of contour tracking.

Conclusions:

• The proposed deform PF-MT algorithm offers a computationally efficient solution for deforming object contour tracking.
• This method significantly advances the state-of-the-art in handling complex deformations and image quality issues.
• The findings enable more robust object tracking in real-world scenarios.