Description

Visual object tracking, i.e., consistently inferring the motion of a target from image sequences, is a must-have component to bridge low-level image processing and high-level video analysis, which gains great popularity due to its applications in diverse areas, such as human-computer interaction, security video surveillance, medical image processing, and robotics. In this book, we focus on how to enhance the generality and reliability of object-level tracking given no prior knowledge about targets. We propose two novel ideas: context-aware and attentional tracking, where the tracker discovers some auxiliary objects that have short-term motion correlations with the target as the spatial contexts, or augments the observation models by selectively attending discriminative regions inside the target, or adaptively tuning the feature granularity and model elasticity. These approaches achieve promising results on challenging real-world videos. The book sheds some light on recent progress of visual tracking and should be useful to professionals in computer vision research, or anyone else who may be considering utilizing visual tracking in intelligent video analysis systems.