Modeling pixel process with scale invariant local patterns for background subtraction in complex scenes

By Shengcai Liao, Guoying Zhao, Vili Kellokumpu, Matti Pietikäinen, S. Li. 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010

405 citations recorded by the source corpus.

Background modeling plays an important role in video surveillance, yet in complex scenes it is still a challenging problem. Among many difficulties, problems caused by illumination variations and dynamic backgrounds are the key aspects. In this work, we develop an efficient background subtraction framework to tackle these problems. First, we propose a scale invariant local ternary pattern operator, and show that it is effective for handling illumination variations, especially for moving soft shadows. Second, we propose a pattern kernel density estimation technique to effectively model the probability distribution of local patterns in the pixel process, which utilizes only one single LBP-like pattern instead of histogram as feature. Third, we develop multimodal background models with the above techniques and a multiscale fusion scheme for handling complex dynamic backgrounds. Exhaustive experimental evaluations on complex scenes show that the proposed method is fast and effective, achieving more than 10% improvement in accuracy compared over existing state-of-the-art algorithms.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/2a29ad33d8a878554008e3f0cad7d3607451d16f

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

1. Yi Yang
2. Xiaogang Wang
3. Qi Tian
4. Liang Zheng
5. Luc Van Gool
6. G. Wang
7. Dacheng Tao
8. Ming-Hsuan Yang

Citations in the corpus

• Similarity Learning with Spatial Constraints for Person Re-identification (2016, background): “Among them, HSV(1) and LAB(1) are 8×8×8 joint histograms, and HSV(2) and LAB(2) are 48 bin concatenated histograms with each channel having 16 bins, HOG[7] and SILPT[20] are texture descriptors.”
• Person re-identification by Local Maximal Occurrence representation and metric learning (2014, methodology): “Within each subwindow, we extract two scales of SILTP histograms (SILTP 0.3 4,3 and SILTP 0.3 4,5 ), and an 8 × 8 × 8-bin joint HSV histogram.”
• A Siamese Long Short-Term Memory Architecture for Human Re-identification (2016, methodology): “– Local Maximal Occurrence (LOMO): To extract the LOMO features, first, Color histogram feature and SILTP features [34] are extracted over 10 × 10 blocks with an overlap of 5 pixels.”
• Ensemble Video Object Cut in Highly Dynamic Scenes (2013, methodology): “Table 1 shows the F -score results on the Combination dataset in comparison with (a) MM03, the Bayes model approach by Li et al. [12]; (b) LBP-P: the pixelwise LBP histogram based approach by Heikkil et al. [8]; and (c) PKDE: the pattern kernel density estimation method by Liao [13] with different p”
• SuBSENSE: A Universal Change Detection Method With Local Adaptive Sensitivity (2015, background): “Another approach was suggested in [35] that does not require histograms: Scale-Invariant Local Ternary Patterns (SILTP) can be used directly at the pixel level to detect local changes”
• Gated Siamese Convolutional Neural Network Architecture for Human Re-identification (2016, neutral): “Scale Invariant Feature Transforms [28,56,57], Scale Invariant Local Ternary Patterns [24, 26], Local Binary Patterns [31, 49], Color Histograms [24,49,56,57] or Color Names [51,60] etc. are the basis of the majority of these feature representations developed for human re-identification.”
• Foreground object detection in highly dynamic scenes using saliency (2014, methodology): “Table 2 shows the accuracy of empty frame detection on this dataset in comparison with (1) PKDE: the pattern kernel density estimation method [12] and (2) EVOC: the ensemble video object cut approach [4].”
• Local binary features for texture classification: Taxonomy and experimental study (2017, background): “In [4] the SILBP was proposed to deal with gray scale intensity changes in complex backgrounds, but again at high feature dimensionality.”