Robust online visual tracking

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01front.pdf133.9 kB
Adobe PDF
Thumbnail02whole.pdf11.22 MB
Adobe PDF
Visual tracking plays a key role in many computer vision systems. In this thesis, we study online visual object tracking and try to tackle challenges that present in practical tracking scenarios. Motivated by different challenges, several robust online visual trackers have been developed by taking advantage of advanced techniques from machine learning and computer vision. In particular, we propose a robust distracter-resistant tracking approach by learning a discriminative metric to handle distracter problem. The proposed metric is elaborately designed for the tracking problem by forming a margin objective function which systematically includes distance margin maximization, reconstruction error constraint, and similarity propagation techniques. The distance metric obtained helps to preserve the most discriminative information to separate the target from distracters while ensuring the stability of the optimal metric. To handle background clutter problem and achieve better tracking performance, we develop a tracker using an approximate Least Absolute Deviation (LAD)-based multi-task multi-view sparse learning method to enjoy robustness of LAD and take advantage of multiple types of visual features. The proposed method is integrated in a particle filter framework where learning the sparse representation for each view of a single particle is regarded as an individual task. The underlying relationship between tasks across different views and different particles is jointly exploited in a unified robust multi-task formulation based on LAD. In addition, to capture the frequently emerging outlier tasks, we decompose the representation matrix to two collaborative components which enable a more robust and accurate approximation. In addition, a hierarchical appearance representation model is proposed for non-rigid object tracking, based on a graphical model that exploits shared information across multiple quantization levels. The tracker aims to find the most possible position of the target by jointly classifying the pixels and superpixels and obtaining the best configuration across all levels. The motion of the bounding box is taken into consideration, while Online Random Forests are used to provide pixel- and superpixel-level quantizations and progressively updated on-the-fly. Finally, inspired by the well-known Atkinson-Shiffrin Memory Model, we propose MUlti-Store Tracker, a dual-component approach consisting of short- and long-term memory stores to process target appearance memories. A powerful and efficient Integrated Correlation Filter is employed in the short-term store for short-term tracking. The integrated long-term component, which is based on keypoint matching-tracking and RANSAC estimation, can interact with the long-term memory and provide additional information for output control.
Please use this identifier to cite or link to this item: