Thesis Defense - Zeynep Gülbeyaz Demirdağ (MSEE)

Zeynep Gülbeyaz Demirdağ – M.Sc. Electrical and Electronics Engineering

Prof. Dr. Hasan Fehmi Ateş – Advisor

Asst. Prof. Dr. İsmail Aktürk – Co-Advisor

Date: 28.12.2022

Time: 10.00

Location: AB4 - B428 Seminer Odası

“ACCELERATION OF IMAGE PROCESSING MODULES IN WIDE-AREA AERIAL SURVEILLANCE”

Prof. Dr. Hasan Fehmi Ateş, Özyeğin University

Asst. Prof. Dr. İsmail Aktürk, Özyeğin University

Prof. Dr. Hasan Fatih Uğurdağ, Özyeğin University

Prof. Dr. Bahadır K. Güntürk, Istanbul Medipol University

Assoc. Prof. Dr. Dionysis Goularas, Yeditepe University

Abstract:

            Wide Area Aerial Surveillance, WAAS, is a type of surveillance system that refers to the monitoring of large geographical areas in real-time. The WAAS system is designed using multiple cameras or sensors mounted on the unmanned aerial vehicle (UAV). WAAS systems can scan an area of several square kilometers at once. However, the huge amount of data collected by WAAS systems can be challenging to process in real-time on board the aircraft. This study proposes to use graphics processing units (GPU) and multi-core programming techniques to accelerate the performance of four key modules of a WAAS system: image matching/stitching, object detection, tracking, and super-resolution. Image matching/stitching involves registering and combining multiple images captured by cameras on UAV to create a panoramic image, in other words, the mosaic image of the area being monitored. Object detection and tracking modules involve identifying and tracking objects' movements, such as cars, trucks, and people. Finally, the super-resolution module uses computational techniques to enhance the resolution of the images and provides more details on the images. By using the GPU and multi-core programming techniques to accelerate these modules, the WAAS system's speed and efficiency are significantly improved. GPUs are well-suited to this task because they are designed for parallel processing, allowing them to process large amounts of data quickly. The experiments show that using GPUs and multi-core programming techniques can significantly improve the performance of image stitching, object detection, tracking, and super-resolution, making it possible to execute these modules in parallel and process the large amount of data collected from WAAS in real-time. The accelerated modules are tested on Nvidia Jetson AGX Xavier embedded GPU card for challenging test scenarios, demonstrating their potential for real-time surveillance on edge devices.

Bio:

Zeynep Gülbeyaz Demirdağ received her BS degree in Electrical and Electronics Engineering from KTO Karatay University in 2019 and her BS degree in Metallurgical and Material Engineering from Selcuk University in 2015. She has been working as a Graduate Teaching and Research Assistant at Özyeğin University under the supervision of Prof. Dr. Hasan Fehmi Ateş, Assoc. Prof. Dr. İsmail Aktürk and Prof. H. Fatih Uğurdağ since February 2020. Her research focuses on the acceleration of image-processing algorithms and embedded systems.