Tracking Randomly Moving Objects on Edge Box Proposals

Gao Zhu,Fatih Porikli,Hongdong Li

DOI: https://doi.org/10.48550/arXiv.1507.08085

2015-11-30

Abstract:Most tracking-by-detection methods employ a local search window around the predicted object location in the current frame assuming the previous location is accurate, the trajectory is smooth, and the computational capacity permits a search radius that can accommodate the maximum speed yet small enough to reduce mismatches. These, however, may not be valid always, in particular for fast and irregularly moving objects. Here, we present an object tracker that is not limited to a local search window and has ability to probe efficiently the entire frame. Our method generates a small number of "high-quality" proposals by a novel instance-specific objectness measure and evaluates them against the object model that can be adopted from an existing tracking-by-detection approach as a core tracker. During the tracking process, we update the object model concentrating on hard false-positives supplied by the proposals, which help suppressing distractors caused by difficult background clutters, and learn how to re-rank proposals according to the object model. Since we reduce significantly the number of hypotheses the core tracker evaluates, we can use richer object descriptors and stronger detector. Our method outperforms most recent state-of-the-art trackers on popular tracking benchmarks, and provides improved robustness for fast moving objects as well as for ultra low-frame-rate videos.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in the tracking of fast - moving and irregularly - moving targets, traditional detection - based tracking methods perform poorly when dealing with fast - moving or randomly - moving targets because they assume that the target motion is smooth and the search window is limited. Specifically, the paper points out:

1. **Limitations of traditional methods**:
- Most detection - based tracking methods use a local search window around the predicted target position in the current frame, assuming that the position in the previous frame is accurate, the trajectory is smooth, and the computing power allows the search radius to accommodate the maximum speed but is small enough to reduce false matches.
- These assumptions may not hold when dealing with fast - moving and irregularly - moving targets, leading to a decline in tracking performance.

2. **Proposed new method**:
- The paper proposes a new target tracker that is not limited to the local search window and can efficiently detect the entire image.
- This method generates a small number of high - quality candidate boxes (proposals) and uses an instance - specific objectness measure to evaluate these candidate boxes.
- During the tracking process, by focusing on updating hard false - positives in the model, it helps to suppress the interference caused by background clutter and learns how to re - rank the candidate boxes according to the object model.

3. **Main contributions**:
- This method significantly reduces the number of assumptions that the core tracker needs to evaluate, so that more abundant object descriptors and stronger detectors can be used.
- The experimental results show that this method outperforms most of the latest advanced trackers on popular tracking benchmarks, especially showing better robustness when dealing with fast - moving objects and ultra - low - frame - rate videos.

In summary, this paper aims to improve the tracking performance of fast - moving and irregularly - moving targets by improving the candidate box generation and model update mechanisms.