İmge karakteristik tabanlı nesne tasviri, saptama ve gerçek zamanlı takibi

Our thesis is based on image feature extraction, description of objects using extracted features, real time detection and comparison of State of art technics. Using global and local image features, two separate proceedings carried out. First proceeding is based on use of local image features and second one is based on global image features. Local image feature extraction technics such as SIFT and SURF are comparatively inspected and using SIFT a new object description and video object tracking project is carried out. For description, we introduced Generic Points notion extracted with use of different perspective planar transformations. These points are robust against object geometrical deformations than the State of art SIFT. Based on Generic Points, a feature tracking algorithm is designed running over video images. Generic features are integrated with RAMOSAC tracker. To provide real-time efficiency, CUDA GPU implementation of SIFT is used. Global image feature extraction technics such as Hmax and Haar Like Simple Features are inspected. The implementation of early Hmax is inspected and a referential feature extraction algorithm is described. Well known Haar Like Simple Features technic is adopted and used for describing object models. This technic is used with Adaboost classifier for preliminary object detection over on road vehicle video records. Using image analyze technics, a suit of detection algorithm is designed to validate preliminary detections. Using color channels for texture analyze, object symmetric features searched. Using edge level images, horizontal lines are detected inside detected region of interests. Symmetric feature search and prominent horizontal line frequency detection are used for validation. Temporal detection story is used for tracking and validation as well. A new proceeding is carried out which enables active safety for on road vehicle navigation. Keywords : SIFT, SURF, HMAX, Haar-Like Simple Features, Object Description, Object Detection, Real-Time Vehicle Detection and Tracking.