Motion Prediction in Visual Object Tracking
This work addresses motion prediction in visual object tracking for applications like autonomous driving, but it is incremental as it combines standard methods.
The paper tackles the problem of visual object tracking by emphasizing motion prediction, achieving state-of-the-art results with improvements in EAO from 0.472 to 0.505 on VOT-2016 and from 0.410 to 0.431 on VOT-2018, and shows generalizability to video object segmentation.
Visual object tracking (VOT) is an essential component for many applications, such as autonomous driving or assistive robotics. However, recent works tend to develop accurate systems based on more computationally expensive feature extractors for better instance matching. In contrast, this work addresses the importance of motion prediction in VOT. We use an off-the-shelf object detector to obtain instance bounding boxes. Then, a combination of camera motion decouple and Kalman filter is used for state estimation. Although our baseline system is a straightforward combination of standard methods, we obtain state-of-the-art results. Our method establishes new state-of-the-art performance on VOT (VOT-2016 and VOT-2018). Our proposed method improves the EAO on VOT-2016 from 0.472 of prior art to 0.505, from 0.410 to 0.431 on VOT-2018. To show the generalizability, we also test our method on video object segmentation (VOS: DAVIS-2016 and DAVIS-2017) and observe consistent improvement.