PHD

Burak Gorkemli

Office door number: ENG 229
Phone number:338 1000/Ext: 2640
E-mail: bgorkemli@ku.edu.tr
Thesis Abstract:Congestion control is crucial for deploying video streaming applications safely over the Internet. TCP-Friendly Rate Control (TFRC) seems to be a promising new scheme to implement congestion control for media transport applications. In this PhD study, I will investigate the methods of using TFRC for streaming video applications based on the H.264 (MPEG4 AVC) codec and its scalable video coding (SVC) extensions. I will consider self imposed TFRC implementations as well as the rate control approach imposed by the emerging Datagram Congestion Control Protocol (DCCP). The target of these investigations is to develop a set of procedures to obtain the best video quality under TFRC using novel approaches for sender-side rate control. Furthermore, I will find innovative ways to implement multiple description coding (MDC) with the H.264 SVC in order to stream video over multipath IP and peer-to-peer (P2P) systems in a TCP-friendly manner.

Ferda Ofli

Office door number: ENG 106
Phone number:338 1000/Ext: 2657
E-mail: fofli@ku.edu.tr
Thesis Abstract:Motion has played an important role in computer vision research from the very beginning and it is becoming much more significant as multiple view environments are being introduced into several areas of this research process. One of these areas is devoted to the study of humans, i.e. face and facial expression recognition, gesture recognition, whole-body tracking and gait recognition, or in the more general sense, complete analysis of human activities. In this thesis, we will investigate two of the several scenarios, one of which can be considered at a micro level with respect to the other one that sits at a macro level, for the analysis of human activities. Our macro level task is to build an automated multicamera system for human body motion capture using color markers. The multiview video of a moving actor is acquired using 8 synchronized cameras. Our motion capture method is based on 3D tracking of markers attached to the person's body in the scene, using stereo color information without need for an explicit 3D model. We employ Kalman filtering to track the 3D positions of markers over frames in a robust manner. The resulting set of 3D points is then used to animate a personalized skeleton-based 3D human body model. Our micro level task is to come up with a framework for joint analysis of facial gestures and speech prosody patterns of a speaker towards automatic realistic synthesis of facial gestures from speech prosody. The set of facial gestures will consist of the movements of eyes, eyebrows, eyelids, and lips of the person. The analysis process aims to "learn" both elementary prosody and facial gesture patterns for a particular speaker, as well as the correlations between these facial gestures and prosody patterns from a training stereo video sequence. HMMs will be employed to find out the correlation metric and to determine the audio-visual.

MS

Yasemin Demir

Office door number: ENG 106
Phone number:338 1000/Ext: 2657
E-mail: ydemir@ku.edu.tr
Thesis Abstract: Detection and Analysis of Unusual Motion Activity from Multiple Cameras
The goal of my research is multiple camera motion analysis for detection of unusual activities of objects including humans. Trajectories and features of objects and interactions between them shall be extracted from each camera. I will use computer vision and pattern analysis techniques for extraction of compact and discriminative model of motion patterns and machine learningtechniques for classifying this time varying feature data as an activity type. Understanding activity, which stands as the last step of the study, covers classification of human motion patterns and generation of a high-level description of actions and interactions. At the end an automatic visual surveillance system with statistical descriptions of typical activity patterns shall be provided, unusual events shall be detected by spotting activities that are very different from normal patterns. Unusual interactions between objects would be tracked, detected and modeled.

Berkin Abanoz

Office door number: ENG 106
Phone number:338 1000/Ext: 2657
E-mail: tabanoz@ku.edu.tr
Thesis Abstract:Rate Distortion Optimization in Monocular and Multiview Videos
Rate Distortion optimization is a key point in video coding. The interest of my research is to do rate distortion optimization while using scalable video coding first with monocular videos than with multiview videos. In other words, the purpose is to do rate allocation between base and enhancement layers.   

Goktug Gurler

Office door number: ENG 229
Phone number:338 1000/Ext: 2640
E-mail: cgurler@ku.edu.tr
Thesis Abstract: Reliable 3D Streaming

 Ozgun Genc

Office door number: ENG 106
Phone number: 338 1000/Ext: 2657
E-mail: ogenc@ku.edu.tr