Applying mean shift, motion information and Kalman filtering approaches to object tracking.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Object tracking in videos is crucial for various applications.
• The mean shift (MS) algorithm, using the Bhattacharyya coefficient, is a prominent method for object localization.
• Standard MS algorithm performance degrades due to background clutter, illumination variations, and occlusions.

Purpose of the Study:

• To enhance the robustness and accuracy of the mean shift algorithm for object tracking.
• To overcome limitations of the MS algorithm, including partial and full occlusions, low saturation, and sudden lighting changes.
• To improve real-time object tracking capabilities.

Main Methods:

• Proposed an improved convex kernel function to handle partial occlusions.
• Integrated motion information with color features to enhance performance under low saturation and varying illumination.
• Utilized a Kalman filter, assuming constant object speed, to address full occlusion scenarios.

Main Results:

• The integrated approach, combining color and motion information, significantly increases the MS algorithm's capability.
• Experimental results demonstrate optimal performance in real-time object tracking.
• The proposed enhancements lead to more reliable tracking compared to the original MS algorithm.

Conclusions:

• The improved MS algorithm offers a robust solution for real-time object tracking in challenging video conditions.
• Simultaneous use of color and motion features, along with Kalman filtering for occlusions, enhances tracking accuracy and reliability.
• This research provides a valuable advancement for video-based object tracking systems.