7651

7123

كارشناسي ارشد

مخابرات

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1391

هشت، ص.: مصور،جدول،نمودار

ص.ع.به فارسي و انگليسي

محمدرضا احمدزاده

26/1/92

مازيار پالهنگ، عبدالرضا ميرزايي

مهندسي برق و كامپيوتر

ID7123

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

78 Object recognition and pose estimation by stereo images Iman Rahimi Pordanjani i rahimipordanjani@ec iut ac ir Date of Submission 2009 05 9 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor M ahmadzadeh@cc iut ac irAbstract Today human has achieved many developments in the field of machine vision One of thesedevelopments is the simultaneous use of several cameras Inspired by the human visual system researchersusually use a two cameras system called stereo system Algorithms and methods for the stereo system usesthe images of the two cameras and the results achieved by the simultaneous processing of two images These processes include various stages during its previous from pre calibration of the camera to the finaloutcome such as distance measurement or three dimensional reconstruction Stereo system encompasses awide range of applications such as building a three dimensional map the location and distance of objects robot navigation grasping by a robot and so In this study we represent an object recognition and poseestimation system based on stereo images that has many applications for mobile robots Stereo systemperforms localization more accurately than mono camera system by using information of two images Inthis system we use SIFT local features SIFT feature is useful tool for object recognition in differentsituations because of invariance to image transforms such as rotation translation and scale changes Alsothis feature is robust to illumination changes and occlusion of objects in images therefor it is suitable forrecognition in different illumination conditions and occlusion scenes One of the most important steps inthese systems is descriptors vector matching with the set of vectors in database to find the correspondingvector or nearest vectors Here we are trying to add parameters to the features to speed up matching stepwithout loss of accuracy that is very important in real time applications This is done by adding parametersto the original feature and the speed of matching step increase in average double the quickest method Therecognition system includes two steps model generation and recognition In both steps we use proposedmethod to speed up feature matching by suitable precision The system has been tested in various modesand system accuracy is calculated for each mode It makes locating objects a few millimeters of errorKeywords 1 Object recognition 2 pose estimation 3 stereo images 4 SIFT features 5 featurematching

محمدرضا احمدزاده

مازيار پالهنگ، عبدالرضا ميرزايي