Real-Time Tracking of Single and Multiple Objects from Depth-Colour Imagery Using 3D Signed Distance Functions.

Area of Science:

• Computer Vision
• Robotics
• 3D Perception

Background:

• Accurate real-time multi-object tracking is crucial for robotics and augmented reality.
• Existing methods often struggle with occlusions and require complex collision detection modules.

Purpose of the Study:

• To develop a unified probabilistic framework for real-time multi-object tracking using RGB-D imagery.
• To represent object shapes implicitly and handle object interactions robustly.

Main Methods:

• Utilizing 3D signed distance functions for implicit object shape representation.
• Developing probabilistic generative models for RGB-D data.
• Formulating tracking as a maximum a posteriori problem.
• Generalizing single-object tracking to multiple objects via shape unions.

Main Results:

• Demonstrated robust real-time tracking of multiple rigid 3D objects.
• The proposed model inherently accounts for object similarity and proximity.
• Eliminated the need for explicit collision detection mechanisms.

Conclusions:

• The novel probabilistic framework offers a robust and efficient solution for multi-object tracking.
• Implicit shape representation and generative models provide a unified approach to handling object interactions.