Saliency-based visual tracking using correlation filters for surveillance applications

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Elektrik ve Elektronik Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2014

Öğrenci: EMRE TUNALI

Danışman: ABDULLAH AYDIN ALATAN

Özet:

In recent years intelligent transportation systems (ITS) have been an active research area in computer vision. One of the main goals of ITS is producing systems to guide surveillance operators and reduce human resources for observing hundreds of cameras in urban traffic surveillance. Thus, this thesis is devoted to realization of low level tasks, target detection and tracking, for an autonomous video surveillance system. The initial step of the proposed system is moving object detection which is utilized based on a recently proposed Self Adaptive Gaussian Mixture Model technique. The resulting moving object mask is further enhanced via post processing steps containing morphological operations and shadow/highlight removal. Benefiting from this enhanced binary mask, track initialization is achieved for each detected moving blob entering to the scene and a track is intended to be maintained until the target leaves the scene. For target tracking, multiple model visual tracking methodology is proposed based on correlation filters. Moreover, in order to adjust tracking parameters online and provide high level tasks with extra information together, a target bounding box generation methodology which is capable of target silhouette extraction is proposed based on temporal consistency of the saliency map of tracking window. The proposed algorithm is tested on synthetic as well as real data and based on these experimental results, it can be concluded that it yields competitive tracking results in real life scenarios.