Derin öğrenme tabanlı çoklu nesne takipçilerinin değerlendirilmesi

Multiple object tracking (MOT) is a significant problem in the computer vision community due to its applications, including but not limited to, surveillance and emerging autonomous vehicles. The difficulties of this problem lie in several challenges, such as frequent occlusion, interaction, intra-class variations, in-and-out objects, etc. Recently, deep learning MOT methods confront these challenges effectively. State-of-the-art deep learning (DL) trackers pipeline consists of two stages, i.e., appearance handling, which includes object detection and feature extraction, and grouping step, which performs affinity computation and data association. One of the main concerns of this thesis is to investigate how DL was employed in each one of these stages. In addition to that, we have experimented with different performance-enhancing methods, the currently top online tracker on the MOTChallenge dataset. Based on the investigation and experiments, we will identify and discuss the significant shortcomings of the current frameworks, providing possible ways to improve it.