A reduced order extended Kalman filter for sequential images containing a moving object.

Area of Science:

• Image Processing
• Computer Vision
• Signal Processing

Background:

• Sequential image analysis often suffers from noise, hindering accurate object tracking and velocity estimation.
• Traditional methods may struggle with low signal-to-noise ratio (SNR) conditions and dynamic object changes.

Purpose of the Study:

• To develop an efficient algorithm for noise reduction and velocity estimation of moving objects in image sequences.
• To introduce a computationally tractable approximation of the Extended Kalman Filter (EKF) suitable for practical applications.

Main Methods:

• The Parallel Extended Kalman Filter (PEKF) was developed, utilizing a bank of third-order EKFs operating on Fourier coefficients.
• A finite impulse response filter was integrated following the EKF bank for enhanced performance.
• The algorithm was designed to model and track slow variations in object velocity.

Main Results:

• The PEKF demonstrated convergence to an optimal algorithm under specific conditions (zero velocity estimation errors).
• Effective performance was shown even in very low SNR image sequences.
• The PEKF's ability to track slow object changes and velocity variations was highlighted.

Conclusions:

• The PEKF provides an efficient and robust method for noise reduction and velocity estimation in challenging image conditions.
• This approach is suitable for applications requiring tracking of dynamic objects and their velocities.
• The PEKF offers advantages over existing frequency domain algorithms for velocity estimation.