We propose a method combine the object detect and shadow removal. Object detection includes two major part of the detect policy, temporal information and spatial information. We design a novel framework, the spatiotemporal background extractor (SBE) that combine the temporal information and spatial information. Two main components are proposed: the background extractor (BE) and the background gradient extractor (BGE). The BE detects the foreground and eliminates the noise in background.
Only use one frame to construct the BE which is based on a single-layer Codebook model and the spatial information is propagated from the adjacent neighbors. The BE can efficiently solve the most challenges of object detection such as dynamic background and the sudden changes of illumination. However the policy of propagation makes the foreground more incompleteness. Therefore, the propagation is forbidden base on the result of BGE that keeps the completeness of foreground. The BGE and BE is constructed synchronously. To construct the BGE, the stable background gradient is used which is used to find foreground gradient of the current frame. In order to handle the problem of shadow removal in visual surveillance. We propose a method combine the advantage of different features of shadow that can remove the shadow base on the properties of each scene. We combine the features, chromaticity, physical properties and texture, from the ground truth. Using the combined data to train the classifier which can remove the shadow from the result of object detection. We choose Random Forest algorithm to train the shadow classifier model because the policy of Random Forest is more suitable for properties of the feature. After training the Random Forest classifier, we use this classifier to remove the shadow from the result of SBE.
In the experimental results of object detection, the result on a set of background modeling databases, Perception and Wallflower. We analyze the performance of the proposed method (SBE) to compare with GMM, Codebook, and ViBe. The total error of the proposed method is the least, and the performance of solving the dynamic background and the sudden changes of illumination are greater than others. In the experimental result of shadow removal, the performance of combine different feature is better than use the single feature. Comparing the method of feature-based and our method, the Random Forest classifier has the higher accuracy than SVM classifier. The result of the proposed method has the similar accuracy with feature-based using feature of texture and better performance than other features. Compare to the method of feature-based which is sensitive to the threshold of each feature, our method can train the model that suitable for each scene by it’s properties.
The result of spatiotemporal background extractor can handle the dynamic background and sudden change of illumination, the foreground has some error region cause by two-way propagation. Although we proposed the forbidden propagation policy, it still has some region of foreground classified to background. We want to use color information about neighbor to solve this problem to get the more complete foreground.
In our proposed work, we use three feature, chromaticity, physical property and texture to train the shadow model. We want to combine other feature do not use the color information and gradient information to improve the performance of shadow removal.
REFERENCE
[1] L. Liyuan, H. Weimin, I. Y. H. Gu, and T. Qi. Statistical modeling of complex backgrounds for foreground object detection. IEEE Transactions on Image Processing, 13(11):1459-1472, 2004.
[2] K. Kim, T. H. Chalidabhongse, D. Harwood, and L. Davis. Real-time foreground–background segmentation using codebook model. Real-time imaging, 11(3):172-185, 2005.
[3] K. Kim, T. H. Chalidabhongse, D. Harwood, and L. Davis. Background modeling and subtraction by codebook construction. In Proceeding of the International Conference on Image Processing (ICIP’04), 3061-3064, 2004.
[4] O. Barnich and M. Van Droogenbroeck. ViBe: A universal background subtraction algorithm for video sequences. IEEE Transactions on Image Processing, 20(6):1709-1724, 2011.
[5] Wenping Wu, Jing Shao, and Wei Guo. Moving-object Detection Based on Shadow Removal and Prospect Reconstruction. Lecture Note in information Technology, vol. 12, 2012.
[6] R. Cucchiara, C. Grana, M. Piccardi, A. Prati, “Detecting moving objects, ghosts, and shadows in video streams,” IEEE Transactions on Pattern Analysis and Machine Intelligence Vol. 25, No.10, pp. 1337-1342, 2003.
[7] J. B. Huang, C. S. Chen, “Moving cast shadow detection using physics-based features,” in Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR’09), pp. 2310-2317, 2009.
[8] A. Sanin, C. Sanderson, B. Lovell, “Improved shadow removal for robust person
tracking in surveillance scenarios,” in Proc. of ICPR’10, pp. 141-144, 2010.
[9] Shih-Chieh Wang, Te-Feng Su and Shang-Hong Lai”Detecting Moving Object From Dynamic Background with Shadow Removal”ICASSP2011 IEEE.
[10] A. Sanin , C. Sanderson , Brian C. Lovell, “Shadow detection: A survey and comparative evaluation of recent methods,” Pattern Recognition, Vol. 45, No. 4, pp.
1684-1695, 2012.
[11] A. Amato, I. Huerta, Mikhail G. Mozerov, F. Xavier Roca ,Jordi Gonzàlez,
“Moving Cast Shadows Detection Methods for Video Surveillance Applications,”
Augmented Vision and Reality, Vol 6, pp 23-47. 2014.
[12] Sohail Nadimi, Bir Bhanu, Physical Models for Moving Shadow and Object Detection in Video. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 26, NO. 8, AUGUST, 2004.
[13] Chia-Jung Chang, Wen-Fong Hu, Hsieh, Jun-Wei, Yung-Sheng Chen,
"Shadow elimination for effective moving object detection with Gaussian models", Pattern Recognition, 2002. Proceedings. 16th International Conference on, On page(s): 540 - 543 vol.2 Volume: 2, 2002
[14] A. Sanin, C. Sanderson, B. Lovell, “Improved shadow removal for robust person tracking in surveillance scenarios,” in Proc. of ICPR’10, pp. 141-144, 2010.
[15] K. Toyama, J. Krumm, B. Brumitt, and B. Meyers. Wallflower: Principles and practice of background maintenance. In Proceedings of the 7th IEEE International Conference on Computer Vision (ICCV’99), 255-261, 1999.
[16] L. Liyuan, H. Weimin, I. Y. H. Gu, and T. Qi. Statistical modeling of complex backgrounds for foreground object detection. IEEE Transactions on Image Processing, 13(11):1459-1472, 2004.
[17] A. Prati, I. Mikic, M. Trivedi, R. Cucchiara, Detecting moving shadows:
algorithms and evaluation, IEEE Transactions on Pattern Analysis and Machine Intelligence 25 (7) (2003) 918–923.