Tracking using Numerous Anchor Points

Abstract

In this paper, an online adaptive model-free tracker is proposed to track single objects in video sequences to deal with real-world tracking challenges like low-resolution, object deformation, occlusion and motion blur. The novelty lies in the construction of a strong appearance model that captures features from the initialized bounding box and then are assembled into anchor point features. These features memorize the global pattern of the object and have an internal star graphlike structure. These features are unique and flexible and helps tracking generic and deformable objects with no limitation on specific objects. In addition, the relevance of each feature is evaluated online using short-term consistency and long-term consistency. These parameters are adapted to retain consistent features that vote for the object location and that deal with outliers for long-term tracking scenarios. Additionally, voting in a Gaussian manner helps in tackling inherent noise of the tracking system and in accurate object localization. Furthermore, the proposed tracker uses pairwise distance measure to cope with scale variations and combines pixel-level binary features and global weighted color features for model update. Finally, experimental results on a visual tracking benchmark dataset are presented to demonstrate the effectiveness and competitiveness of the proposed tracker.

Robust face tracking using multiple appearance models and graph relational learning

“…It has been shown in [15] and [16] that structure can be a powerful appearance representation for visual object tracking. Whereas in [14], it has been shown that the structure of an object can help to tackle occlusion. Our motivation for using structure is inspired by the idea that by exploring the intrinsic structure of a target face may help to discover a particular pattern of a face of interest.…”

“…To adapt to the scale variation of the face, we use the same strategy, as used in [14]. The authors utilized pairwise distances between matched keypoints between consecutive frames to tackle scale change.…”

“…The proposed method is validated on OTB benchmark [22] for One-Pass Evaluation (OPE). The selected state-of-the-art trackers used for comparison are: Struck [5], TLD [3], KCF [7], MIL [6], CMT [16], TUNA [14] and Boosting [4]. 15 video sequences from the benchmark containing faces are chosen for evaluation.…”

“…Scale variation and rotation: Together with the keypoint scale adaptation strategy from [14], and scale and aspect ratio adaptation from a face detector, the tracker performs well in tackling scale variation of the face, which is a common phenomena during object tracking. As long as the face remains partially or fully visible during in plane rotation and out-of-plane rotation, all the appearance models namely, GRM, ICM, and the BDM models contribute in face localization by maximizing the fusion score for all the face candidates.…”

“…They use online samples for updating the appearance of the face, and employ forgetting factor for adapting the appearance model. Related to our work are object trackers [14], [15] and [16] that utilize structure of the object as the appearance representation for tracking.…”