Thumbnail
Access Restriction
Open

Author A. Oostdijk, A. ♦ M. Van Persie, A. ♦ H. H. S. Noorbergen, A. ♦ J. W. Van Rijn, A.
Source CiteSeerX
Content type Text
File Format PDF
Subject Domain (in DDC) Computer science, information & general works ♦ Data processing & computer science
Subject Keyword Medium Level ♦ Relevant First Achievement ♦ Basic Segmentation ♦ Interpretation Process ♦ Fine Level ♦ Object Characteristic ♦ Operator Knowledge ♦ Single Image ♦ High Resolution Dataset ♦ Multiple Image ♦ Multispectral Aerial Photograph ♦ Satellite Image ♦ Autonomous Way Region ♦ Multiple Optical Image ♦ Contextual Rule ♦ Multi Scale ♦ Relational Knowledge ♦ Security Application ♦ Detected Road Segment ♦ Image Resolution ♦ Optical Road ♦ Type Matching ♦ Accurate Correction ♦ Demonstration Environment ♦ Object Geometry ♦ Extended Use ♦ Considerable User Interaction ♦ User Interaction ♦ Network Building Step ♦ Cot Software Environment ♦ Change Detection Scheme ♦ Classification Result ♦ Detected Vehicle ♦ Course Level ♦ Target Extraction ♦ Image Interpretation
Abstract In the framework of defence and security applications, NLR has built a demonstration environment and did experiments to use object based image interpretation for the detection of roads and vehicles in single or multiple optical images. The usage of an object based approach provides the ability to include part of the operator knowledge into the software based interpretation process. The method of working for the optical road and target extraction in a single image is developed in a multi scale, multi method scheme in which three levels can be recognized. Firstly the course level, in which in an autonomous way Regions Of Interest (ROI’s) are determined. Secondly the medium level, in which basic segmentation and classification is done within the ROI’s, with some user interaction. Thirdly in the fine level, the classification result is refined based on the knowledge on object characteristics and contextual rules. On this level, considerable user interaction is required. For detected vehicles a form of type matching can be done depending on the image resolution. For detected road segments, cleaning and network building steps are applied. If multiple images are available, a change detection scheme is added at the medium level. The schemes have been implemented in a COTS software environment. The method has been tested on a high resolution dataset, consisting of multispectral aerial photographs and satellite images. The results show a relevant first achievement, but improvements are required. Options are more accurate correction for the 3D object geometry and more extended use of relational knowledge and probabilities. 1.
Educational Role Student ♦ Teacher
Age Range above 22 year
Educational Use Research
Education Level UG and PG ♦ Career/Technical Study