Finding trajectories of feature points in a monocular image sequence

Area of Science:

• Computer Vision
• Robotics
• Motion Analysis

Background:

• The correspondence problem, identifying points across images, is crucial for motion analysis.
• Existing methods often rely on only two image frames, limiting their effectiveness.
• Exploiting motion smoothness over multiple frames offers a more robust solution.

Purpose of the Study:

• To develop a novel algorithm for solving the correspondence problem using sequential image frames.
• To address the challenge of tracking nonrigid objects with arbitrary motion.
• To handle occlusions effectively in motion analysis.

Main Methods:

• Formulating the correspondence problem as an optimization problem.
• Proposing an iterative algorithm to compute point trajectories in monocular image sequences.
• Modifying the algorithm to accommodate occlusions.

Main Results:

• Demonstrated the algorithm's efficacy in solving the correspondence problem for nonrigid objects.
• Successfully tracked point trajectories using motion smoothness and sequential frames.
• Validated the approach on synthetic, laboratory, and real-world scenes.

Conclusions:

• The proposed iterative algorithm effectively solves the correspondence problem by leveraging motion smoothness.
• The method provides robust point trajectory estimation for nonrigid objects, even with occlusions.
• This approach enhances motion analysis capabilities in computer vision applications.