Object Recognition using Channel-Coded Feature Maps: C++ Implementation Documentation
2008 (English)Report (Other academic)
This report gives an overview and motivates the design of a C++ framework for object recognition using channel-coded feature maps. The code was produced in connection to the work on my PhD thesis Channel-Coded Feature Maps for Object Recognition and Machine Learning. The package contains algorithms ranging from basic image processing routines to specific complex algorithms for creating channel-coded feature maps through piecewise polynomials. Much emphasis has been put in creating a flexible framework using virtual interfaces. This makes it easy e.g.~to switch between different image primitives detectors or learning methods in an object recognizer. Some common design choices include an image class with a convenient but fast pixel access, a configurable assert macro for error handling and a common base class for object ownership management. The main computer vision algorithms are channel-coded feature maps (CCFMs) including their derivatives, single-sided colored lines, object detection using an abstract hypothesize-verify framework and tracking and pose estimation using locally weighted regression and CCFMs. The code is considered as having alpha status at best. It is available under the GNU General Public License (GPL) and is mainly intended for future research on the subject.
Place, publisher, year, edition, pages
Linköping, Sweden: Linköping University, Department of Electrical Engineering , 2008. , 17 p.
LiTH-ISY-R, ISSN 1400-3902 ; 2838
Engineering and Technology
IdentifiersURN: urn:nbn:se:liu:diva-53340ISRN: LITH-ISY-R-2838OAI: oai:DiVA.org:liu-53340DiVA: diva2:288558