Advances in Detection and Classification for Through-the-Wall Radar Imaging.
Technische Universität, Darmstadt
[Ph.D. Thesis], (2010)
Available under Creative Commons Attribution Non-commercial No Derivatives, 2.5.
Download (4MB) | Preview
|Item Type:||Ph.D. Thesis|
|Title:||Advances in Detection and Classification for Through-the-Wall Radar Imaging|
In this PhD thesis the problem of detection and classification of stationary targets in Through-the-Wall Radar Imaging is considered. A multiple-view framework is used in which a 3D scene of interest is imaged from a set of vantage points. By doing so, clutter and noise is strongly suppressed and target detectability increased. In target detection, centralized as well as decentralized frameworks for simultaneous image fusion and detection are examined. The practical case when no prior knowledge on image statistics is available and all inference must be drawn from the data at hand is specifically considered. An adaptive detection scheme is proposed which iteratively adapts in a non-stationary environment. Optimal configurations for this scheme are derived based on morphological operations which allow for automatic and reliable target detection. In a decentralized framework, local decisions are transmitted to a fusion center to compile a global decision. In these scenarios, the concept of confidence information of local decisions is crucial to obtain acceptable detection results. Confidence information is classically based on prior knowledge on either the image statistics or local detector performance which generally are unknown in practice. A novel adaptive fusion scheme based on the bootstrap is proposed to automatically extract confidence information of local decisions given the acquired data at hand. In target classification a general framework consisting of segmentation, feature extraction and target discrimination is proposed. The adaption of all these techniques to the application of Through-the-Wall Radar Imaging is investigated, whereby the focus is set on the feature extraction step. A combination of statistical and geometrical features based on superquadrics is proposed. It is shown that most features depend on system and scene parameters such as system resolution and target distance. Compensation methods to allow for resolution-independent feature extraction are consequently derived. All proposed methods are evaluated using simulated as well as real data measurements obtained from three-dimensional imaging measurements using wideband sum-and-delay beamforming.
|Place of Publication:||Darmstadt|
|Uncontrolled Keywords:||Radar Imaging, Detection, Classification|
|Classification DDC:||600 Technik, Medizin, angewandte Wissenschaften > 620 Ingenieurwissenschaften|
|Divisions:||18 Fachbereich Elektrotechnik und Informationstechnik|
|Date Deposited:||13 Sep 2010 14:01|
|Last Modified:||07 Dec 2012 11:58|
|Referees:||Zoubir, Prof. Dr.- Abdelhak and Amin, Prof. Dr. Moeness|
|Refereed:||20 August 2010|