3199

3133

كارشناسي ارشد

﴿هوش مصنوعي و رباتيك﴾

اصفهان:دانشگاه صنعتي اصفهان،دانشكده برق و كامپيوتر

1384

ده،109، [II] ص:مصور،جدول، نمودار

ص.ع.به فارسي و انگليسي

محمد داور پناه جزي

مازيار پالهنگ

سعيد صدري،لاري

مهندسي برق و كامپيوتر

ID3133

به فارسي و انگليسي:قابل رويت در نسخه ديجيتالي

Abstract Visual surveillance in dynamic scenes especially for humans and vehicles is currently oneof the most active research topics in computer vision The processing framework of visualsurveillance in dynamic scenes includes the following stages modeling of environments detection of motion classification of moving objects tracking understanding and descriptionof behaviors In this thesis we have presented a new hybrid method for vehicle detection tracking and extracting vehicle parameters The proposed hybrid method uses region andobject features in two steps to cope with tracking issues such as appearance disappearance splitting and occlusion in a cluttered background This method works as prediction correctionmechanism that uses region tracker as a predictor for object tracker and object tracker as acorrector that verifies the result of the region predictor corrects its errors and updates itsstates Region tracking is based on a flexible procedure that exploits the region feature vectorsin two steps The first step predicts moving region in the next frame The second step verifiesand refines predicted regions The Region Prediction uses contour model of each regionpartition and moves information to predict region partitions in the next frame These partitionswork as a guide for object segmentation and object tracking in the next frame Imagesegmentation is based on adaptive background model Most of proposed methods for imagesegmentation are not robust against light changes and have problems in this situation Adaptive background is robust against light changes and could be implemented as a real timemethod Object tracking is based on a feature based method and includes three steps In thefirst step the positions of the objects are predicted using Kalman filter Then their existence isverified using a novel feature extraction method and particle swarm optimization as a searchmechanism The second step finds correspondence pairs in the current frame and the nextframe In the third step the corresponding objects are selected in the current frame and in thenext frame After tracking different parameters of vehicle such as speed size position anddirection are extracted Using these parameters we have proposed a new hierarchical structurefor analyzing activities of each vehicle

محمد داور پناه جزي

مازيار پالهنگ

سعيد صدري،لاري