master thesis

SignatureMA 017
AuthorDavid Geier
TitleDevelopment and Evaluation of a Real-time Capable Multiple Hypothesis Tracker
TutorDipl.-Ing. Michael P├Ątzold
ProfessorDr.-Ing. Thomas Sikora
AbstractMulti-object tracking is nowadays deployed in a vast number of scientific disciplines. Even though multi-object tracking has been researched for decades it still remains a challenging problem because of its computational complexity. The aim of this master thesis is to analyze the real-time capabilities of a modern multiple-hypothesis tracker used for tracking persons in (surveillance) videos. For this purpose all required theory is discussed first and afterwards, the implementation created in the scope of this master thesis is presented. What follows is an evaluation which focuses on the speed of the implementation. Potential bottlenecks are analyzed and ideas for future optimizations are given. Supplementary, the tracking quality is analyzed and compared to the one of a global nearest neighbour tracker.
Key wordssingle target tracking, multiple target tracking, multiple hypothesis tracking, implementation, evaluation