Learning-based Tracking of Fast Moving Objects
This addresses a specific challenge in computer vision for applications like surveillance or sports analysis, but it is incremental as it builds on existing deep learning methods.
The paper tackles the problem of tracking fast moving objects (FMOs) that appear blurred in videos by proposing a learning-based tracking-by-segmentation approach, achieving near-realtime performance on real-world sequences.
Tracking fast moving objects, which appear as blurred streaks in video sequences, is a difficult task for standard trackers as the object position does not overlap in consecutive video frames and texture information of the objects is blurred. Up-to-date approaches tuned for this task are based on background subtraction with static background and slow deblurring algorithms. In this paper, we present a tracking-by-segmentation approach implemented using state-of-the-art deep learning methods that performs near-realtime tracking on real-world video sequences. We implemented a physically plausible FMO sequence generator to be a robust foundation for our training pipeline and demonstrate the ease of fast generator and network adaptation for different FMO scenarios in terms of foreground variations.