Adaptive Color Attributes for Real-Time Visual Tracking

By Martin Danelljan, Fahad Shahbaz Khan, Michael Felsberg, Joost van de Weijer. 2014 IEEE Conference on Computer Vision and Pattern Recognition, 2014

1,507 citations recorded by the source corpus.

Visual tracking is a challenging problem in computer vision. Most state-of-the-art visual trackers either rely on luminance information or use simple color representations for image description. Contrary to visual tracking, for object recognition and detection, sophisticated color features when combined with luminance have shown to provide excellent performance. Due to the complexity of the tracking problem, the desired color feature should be computationally efficient, and possess a certain amount of photometric invariance while maintaining high discriminative power. This paper investigates the contribution of color in a tracking-by-detection framework. Our results suggest that color attributes provides superior performance for visual tracking. We further propose an adaptive low-dimensional variant of color attributes. Both quantitative and attribute-based evaluations are performed on 41 challenging benchmark color sequences. The proposed approach improves the baseline intensity-based tracker by 24 % in median distance precision. Furthermore, we show that our approach outperforms state-of-the-art tracking methods while running at more than 100 frames per second.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/f5dbe4550d24d5374d9e10fce44a35b105c7ee07

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

The Visual Object Tracking VOT2017 Challenge Results (2017, background): “Standard Discriminative Correlation Filter (DCF) based trackers such as [13, 16, 26] suffer from the inherent periodic assumption when using circular correlation.”
Learning Spatial-Temporal Regularized Correlation Filters for Visual Tracking (2018, methodology): “And more discriminative features are widely used, such as HOG [8], color names (CN) [14] and deep CNN features [28, 34].”
A Scale Adaptive Kernel Correlation Filter Tracker with Feature Integration (2014, methodology): “Our proposed approach is closely related to the correlation filter-based trackers [14] [4] [12] [7] [6], which adopt the correlation filter in traditional signal processing technique into the tracking applications.”
Accurate Scale Estimation for Robust Visual Tracking (2014, methodology): “The discriminative correlation filters described in section 2 have recently been extended to multi-dimensional features for a variety of applications, including visual tracking [4, 13], object detection [8, 12] and object alignment [2].”
CREST: Convolutional Residual Learning for Visual Tracking (2017, background): “The examples include kernelized correlation filters [17], multiple dimensional features [12, 18], context learning [49], scale estimation [7], re-detection [31], subspace learning [28], shortterm and long-term memory [20], reliable collection [27] and spatial regularization [9].”
Using labeled data to evaluate change detectors in a multivariate streaming environment (2009, neutral): “The proposed network is implemented by Pytorch and run on the workstation with Intel Core CPU i5-7500 (3.40GHz) and Where N represents the number of pixels, i represents the index of pixels, M represents the number of categories, c represents the index of categories, y ic is the label of pixel i ,…”
The Visual Object Tracking VOT2015 Challenge Results (2015, methodology): “The feature used in original KCF is extended to a combination of HOG, intensity, and colour naming similarly to [13, 44], and the robust model updating scheme in [13] is also adopted.”
Understanding and Diagnosing Visual Tracking Systems (2015, methodology): “Some recent methods [10, 25] demonstrated notable performance with carefully designed color features.”
Long-term correlation tracking (2015, methodology): “A typical tracker [3, 10, 6, 28, 5] based on correlation filters models the appearance of a target object using a filter w trained on an image patch x of M ×N pixels, where all the circular shifts of xm,n, (m,n) ∈ {0, 1, .”
RGB-T Object Tracking: Benchmark and Baseline (2018, background): “More and more discriminative features are utilized in tracking for robust representation of the objects, such as HOG [44], color names [45] and deep features [46, 47].”
The Visual Object Tracking VOT2014 Challenge Results (2014, positive): “CAT extends CSK tracker [26] to multi-channel colour features and it also augments CAT by including texture features and shape features.”
Struck: Structured Output Tracking with Kernels (2016, neutral): “In this paper, we take a different approach and frame the overall tracking problem as one of structured output prediction, in which the task is to directly predict the change in object configuration between frames.”
Hierarchical Convolutional Features for Visual Tracking (2015, methodology): “of this connection, we learn one adaptive correlation filter [3, 16, 7, 35, 6, 17] over features extracted from each CNN layer and use these multi-level correlation response maps to collaboratively infer the target location.”
LaSOT: A High-quality Large-scale Single Object Tracking Benchmark (2020, background): “, ECO HC [14], DSST [16], CN [20], CSK [38], KCF [39], fDSST [17], SAMF [55], SCT4 [10], STC [90] and Staple [5]) or deep features (e.”
Discriminative Correlation Filter Tracker with Channel and Spatial Reliability (2016, neutral): “[12] proposed multi-dimensional color attributes and Li and Zhu [31] applied feature combination.”