| The tracking based on the contour of interested objects is an important and valuable basicresearch in the field of computer vision. And the recognition of the objects and the extraction ofthe contour are the key technologies. In order to improve the tracking accuracy, this thesisdesigned a system for objects recognition and contour extraction based on over-segmentation.The main research contents and solved problems are as follows:1. Because there are many objects in the scene, by the construction of the template libraryfor the target objects, the objects can be found efficiently in the video image sequence.And it improves the speed and efficiency for interesting object recognition.2. In this thesis, a method for object recognition was put forwards which combined theFAST (Feature from Accelerated Segment Test), which are strong robust key pointsdetector, with the local descriptor DASIY, which are the rotation of deformation. And itcan improve the accuracy for object recognition and location in the feature matching indifferent scenes with different light intensities.3. Over-segmentation algorithm was introduced to extract the contours of the objects. And,an evaluation method about over-segmentation was put forwards combined theoptimization method and the experimental method. Due to the instead of the ground truth by our real-time boundary extraction, the evaluation method was used to choosethe best segmentation method with the best matching with the object boundary.4. This thesis studied the boundary closure algorithm with the clustering of super-pixels,and put forwards a method based on EGIS for contour extraction. And this method canget an extract the contour for objects correctly.This thesis based on the over-segmentation studied the recognition and contourextraction for interested objects. And it can improve the accuracy for object recognitionand contour extraction. As a result, it can be used in many areas, such as aviation,intelligent monitoring system, tracking system and so on. |
PDF Full Text Request