In this dissertation, I focus on the feature extraction for human detection.
Contributions are given in the proposal of GLBP methods and the improvement of GLBP method as follows.
Gradient Local Binary Patters (GLBP) method is proposed. In GLBP method, gradient information and texture information are intra-combined. Compared with other gradient based methods, GLBP method can get more powerful edge direction information and remove some pixel noise. Compared with other texture based methods, GLBP can give the different weight for each pixel by gradient information to focus the important part of the images. The experiments on INRIA Dataset show that, GLBP Linear Detector gets 92% Hit Rate in 10-4 False Positive Per Window (FPPW), where HOG gets 87%, HOT gets 89% and S-LBP gets 91.7%.
Two additional parts, Pattern Cutting and New Gradient Formulas are proposed for GLBP method. In GLBP method, 56 Uniform Patterns are selected from 256 (28) Local Binary Patterns to reduce the feature vector length. In these Uniform Patterns,
other 200 Non-Uniform Patterns is ignored. So Patterns Cutting method is proposed to cut some non-uniform patterns to several uniform patterns to reuse the lost information. In the computing of gradient values, I always choose 4 nearby pixel for computing. In GLBP, the local binary code of nearby 8 pixels is calculated. That means I can select the best pixels from these 8 pixels for gradient value computing.
New Gradient Formulas are proposed by this purpose. From the experiments of INRIA data, the RBF based GLBP detector is developed a lot by adding these two additional parts. GLBP RBF Detector gets 96.7% Hit Rate in 10-4 False Positive Per Window (FPPW), where HOG gets 91%, HOT gets 94% and S-LBP gets 92%.
Multi-Level GLBP method, Multi-Block GLBP are proposed to develop the detection rate by Multi-Scale Methods. In Multi-Level GLBP method, some bigger size blocks are added to get the block feature vectors and combine the final feature
Conclusion
vector. Only the computations of voting in new blocks are added in the feature extraction. In Multi-Block GLBP method, some same size blocks are added to get the block feature vectors in scaled images. Not only the voting with blocks is added, but also the feature extraction on scaled images are added which costs more computations and times. The result shows that Multi-Block GLBP method can get better detection rate than Multi-Level GLBP method. Sub-windows Re-Decision Method is proposed to improve the detection rate by nearby sub-windows results in merge part. Some wrong detection and false detection can be corrected by this method.
Reference
[1]
Conference on Computer Vision and Pattern Recognition (CVPR), Vol.1, pp.886-893, 2005.
[2] human detection using a
Pattern Recognition, pp.1491-1498, New York, 2006.
[3]
attern Anal. Mach. Intell., vol.30, no.10, pp.1713-1727, Oct.
2008.
[4]
pp.1-8, 2008.
[5] , In
International Conference on Acoustics, Speech and Signal Processing, pp.2186-2189, 2010.
[6]
IEEE Conference on Computer Vision and Pattern Recognition, Vol.1, pp.511-518, 2001.
[7]
pp.90-97, 2005.
[8] B.Wu and R.Nevatia, Detection and tracking of Multiple, partially occluded humans by bayesian combination of edgelet based part detectors , IJCV, pp.247-266, 2007.
Reference
[9] .Mikolajczyk, et al., Human detection based on a probabilistic assembly of robust part detector, inECCV, 2004, pp. 69-82.
[10] HOG-LBP Human Detector with Partial
, pp.32-39, 2009.
[11]
pp.1030-1037, 2010.
[12] N.Jiang, J.Xu and S.Goto, Pedestrian detection using Gradient Local Binary Patterns , IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, Vol.E95-A, No.8, August 2012.
[13] Ning Jiang, Jiu Xu, Wenxin Yu and Satoshi Goto,
2013 IEEE International Symposium on Circuits and Systems (ISCAS), pp.978-981, Beijing, China, May 2013.
[14] A.Maji, A.Berg, and J.Malik, Classification using intersection kernel support vector machines is efficient , Proc. International Conf. on Computer Vision and Pattern Recognition, pp. 1-8, 2008.
[15] Z.wei, G.Zelinsky, and D.Samaras, Real-time accurate object detection using multiple resolutions , Proc. International Conf. on Computer Vision, pp.1-8, 2007.
[16] S.Tang and S.Goto, Multi scare block histogram of template feature for pedestrian detection , Proc. International Conf. on Image Processing, pp.3493-3496, 2010.
[17] s
in CVPR 2005, pp.878-885, 2005.
[18]
[19] -in CVPR 1999, pp.87-93, 1999.
[20] -template
, pp.1-8, 2007.
[21] Z.Lin and S.Larry, A Pose-Invariant Descriptor for Human Detection and Segmentation , in ECCV, pp.423-436, 2008.
[22]
-vol.60, pp.91-110, 2004.
[23] B.Scholkopf and A.Smola, Learning with Kernels Support Vector Machines, Regularization, Optimization and Beyond, published by the MIT Press, 2002.
[24] J.Friedm
-407, 2000.
[25] G.Zhao
patterns with an application to facial expression Intell., 29(6):915-928, 2007.
[26] Ning Jiang, Yijun Lu, Shaopeng Tang and Satoshi Goto, of Separate Haar Features f -CSCC, pp.128-131, Pattiya, Thailand, May 2010.
[27] T.Ojala, M. Pietikäinen a
Elsevier Editorial System for Pattern Recognition, 29(1):51-59, 1996.
[28] y
28(12):2037-2041, 2006.
Reference
[29]
n and Pattern Recognition, pp. 797 804, 2004.
[30] C.C.Chang, C.J.Lin, LibSVM Software [Online], http://www.csie.
ntu.edu.tw/~cjlin/libsvm/
[31]
In Proc.of IEEE Conf.on Computer Vision and pattern Recognition, pages 639-646, 2004.
[32]
IEEE Transactions on Pattern Analysis and Machine Intelligence, 20:22-38, 1998.
[33] Masayuki Hiromoto, Kentaro Nakahara, Hiroki Sugano, Yukihiro Nakamura, Ryusuke Miyamoto, A Specialized Processor Suitable for AdaBoost-Based Detection with Haar-like Features, CVPR, pp.1-8, 2007.
[34] Ning Jiang Cascade
, 2011 IEEE 7th International Colloquium on Signal Processing and its Applications (CSPA), Penang, Malaysia, pp.155-158, March 2011.
[35] Ning Jiang
using A Multi- 2010 International
Symposium on Intelligent Signal Processing and Communication Systems (ISPACS), pp.1-4, Chengdu, China, December 2010.
[36] INRIA Dataset [Online], Avilable: http://lear.inrialpes.fr/data [37] Open Computer Vision Library (Opencv), by Intel Corporation.
[38] Cheng, Yizong
1995.
[39] Z.Wei, et al., Real-time accurate object detection using multiple resolutions, in ICCV, pp.1-8, 2007.
[40] Shaopeng.T and Satoshi.G, Multi scale block histogram of template feature for pedestrian detection , in ICIP, pp.3493-3496, HongKong, 2010.
Reference
Publication List
Journal Paper:
[1] Ning Jiang, Jiu Xu, Wenxin Yu -combined feature for IIEEJ Transactions on Image Electronics and Visual Computing Vol.1, No.1, pp.88-96, December 2013.
[2] Jiu Xu, Ning Jiang, Heming Sun, Axel Beaugendre and Satoshi Goto, -time Human Detection Based on Multi-scale Bidirectional Local Template , IIEEJ Transactions on Image Electronics and Visual Computing, Vol.1, No.1, pp.28-37, Dec. 2013.
[3] Jiu Xu, Ning Jiang, and Satoshi Goto, irectional local template patterns: An actions on Fundamentals of Electronics, Communications and Computer Sciences, vol.E96-A, no.6, pp.1204-1213 June 2013.
[4] Ning Jiang, Jiu Xu and
actions on Fundamentals of Electronics, Communications and Computer Sciences, vol.E95-A, no.8, pp.1280-1287, August 2012.