Estimation of object motion parameters from noisy images

Area of Science:

• Computer Vision
• Robotics
• Estimation Theory

Background:

• Estimating object motion from image sequences is crucial for applications in robotics and computer vision.
• Traditional methods often struggle with noisy data and limited correspondence points.

Purpose of the Study:

• To develop a robust approach for estimating rigid body motion parameters (rotation and translation) from noisy image sequences.
• To leverage object dynamics modeling for improved motion estimation accuracy.

Main Methods:

• Utilized recursive and batch techniques to extract motion parameters by modeling object dynamics over time.
• Employed an iterated extended Kalman filter for the recursive estimation of motion for a 2D object in 1D images.
• Investigated noise levels of 5-10% of object image size.

Main Results:

• Demonstrated the ability to achieve significant smoothing using a large number of images.
• Derived approximate Cramer-Rao lower bounds for model parameter estimates.
• Showcased the effectiveness of the iterated extended Kalman filter for recursive motion estimation.

Conclusions:

• The proposed approach provides a viable method for object motion estimation in scenarios with noisy images and limited match points.
• This technique is particularly useful when long image sequences are available.
• The method addresses the central projection model's inherent ambiguity in depth information.