Object Tracking in FPV

There is a need to understand and focus on the problems of Visual Object Tracking in First Person Vision (FPV), which is important in applications such as augmented reality, human-computer interaction, surveillance, and sports analysis. Using TREK-150, a benchmark dataset with 150 richly annotated FPV video sequences, we conducted a thorough analysis, comparing the performance of seven algorithms, including both generic and FPV-specific trackers. Our findings highlight the specific challenges of object tracking in FPV and suggest future research directions. Despite some obstacles, our findings show that trackers can be useful in short-term FPV activities. These algorithms, which have been directed to handle appearance changes, camera movements, and background dynamics, show promise in terms of boosting tracking accuracy. Furthermore including object re-identification skills improves models’ ability to deal with occlusions and reappearance. We recommend the incorporation of FPV-specific tracking algorithms that can adapt to dynamic FPV data. In this regard, deep learning-based tracking approaches have the potential to improve performance, making them useful across a wide range of disciplines. Our research predicts the continuous evolution of general object tracking approaches tailored to the unique problems of First Person Vision.