Adaptive matched filter for implicit-target recognition: application in three-dimensional reconstruction.

Area of Science:

• Optical Engineering
• Computer Vision
• Image Processing

Background:

• Traditional matched filters for optical correlators necessitate precise target shape information, limiting their application scope.
• Unspecified or dynamically changing target appearances pose challenges for conventional filter designs.

Purpose of the Study:

• To design an adaptive correlation filter for optical correlators that operates without explicit target shape knowledge.
• To address the challenge of detecting implicit targets within cluttered backgrounds of unknown statistical properties.

Main Methods:

• The study optimizes a mean-squared-error criterion for adaptive filter design.
• Estimators are proposed to determine the target's region of support and statistical parameters of scene noise.
• The filter's performance is evaluated using detection efficiency and location accuracy metrics.

Main Results:

• The proposed adaptive filter successfully detects implicit targets embedded in cluttered scenes.
• The developed estimators effectively characterize target regions and noise statistics.
• The filter demonstrates robust performance in stereo matching and 3D reconstruction tasks.

Conclusions:

• The adaptive correlation filter offers a viable solution for optical correlation tasks with unspecified target shapes.
• This approach enhances the utility of optical correlators in dynamic and complex environments.
• The method shows promise for improving accuracy in 3D reconstruction and stereo vision applications.