Pedestrian detection based on deformable part model with branch and bound

doi:10.11772/j.issn.1001-9081.2017.07.2003

Journal of Computer Applications ›› 2017, Vol. 37 ›› Issue (7): 2003-2007.DOI: 10.11772/j.issn.1001-9081.2017.07.2003

Previous Articles Next Articles

Pedestrian detection based on deformable part model with branch and bound

CHAI Enhui, ZHI Min

College of Computer and Information Engineering, Inner Mongolia Normal University, Hohhot Nei Mongol 010022, China

Received:2016-12-21 Revised:2017-03-02 Online:2017-07-10 Published:2017-07-18
Supported by:
This work is partially supported by the Research Foundation of Inner Mongolia Normal University (2016ZRYB005).

融合分支定界的可变形部件模型的行人检测

柴恩惠, 智敏

内蒙古师范大学计算机与信息工程学院, 呼和浩特 010022

通讯作者: 智敏
作者简介:柴恩惠(1992-),女,山西原平人,硕士研究生,主要研究方向:视频检索、图像处理;智敏(1972-),女,内蒙古巴林左旗人,教授,博士,主要研究方向:视频检索、图像处理。
基金资助:
内蒙古师范大学科研基金资助项目（2016ZRYB005）。

Abstract

Abstract: The detection accuracy of the Deformable Part Model (DPM) algorithm is high in the field of pedestrian detection, however, in the two steps of feature extraction and pedestrian location, the computation is too large, which leads to the slow detection speed and the deformable part model algorithm can not be used in real time pedestrian detection. To solve the problems, a deformable Part Model with Branch and Bound (BB) algorithm and Cascaded Detection (CD) algorithm (BBCDPM) was proposed. First, the Histogram of Oriented Gradients (HOG) feature was selected to describe human target to generate characteristic pyramid. Then, the deformable part model was modeled, and the Latent Support Vector Machine (LSVM) was used to train the model. In order to increase the accuracy of pedestrian detection, the part model of traditional deformation part model algorithm was increased from 5 to 8 parts. Finally, the cascade detection algorithm was used to simplify detection model, then the maximum value was found by combining with the branch and bound algorithm, and a lot of impossible object assumptions were removed, so the pedestrian target location and detection were completed. The experimental results on INRIA dataset show that, compared with the traditional DPM algorithm, the proposed algorithm improves the accuracy rate by 12 percentage points and significantly accelerates pedestrian detection and recognition.

Key words: Branch and Bound (BB) algorithm, Deformable Part Model (DPM) algorithm, Cascaded Detection (CD) algorithm, Histogram of Oriented Gradients (HOG) feature, characteristic pyramid, Latent Support Vector Machine (LSVM), pedestrian detection

摘要： 针对可变形部件模型（DPM）算法在行人检测领域中的检测精度高，但由于在特征提取和行人定位两步中的计算量过大，导致检测速度过慢而不能应用于实时行人检测的问题，提出了一种融合分支定界算法和级联检测算法的可变形部件模型（BBCDPM）算法。首先，选取梯度方向直方图（HOG）特征作为描述人体目标的特征，从而生成特征金字塔；然后，进行可变形部件模型的建模，并使用隐变量支持向量机（LSVM）对模型进行训练；同时，为了提高行人检测的准确度，将传统可变形部件模型算法中的5个部件模型增加到了8个；最后，在利用了级联检测算法简化检测模型的基础上，结合了分支定界算法寻找最大值，排除大量不可能的对象假设，完成对行人目标的定位和检测。在INRIA数据集上进行了实验，结果表明，与传统DPM算法相比，该算法将准确率提高了12个百分点，且大幅提高了行人检测与识别的速度。

关键词: 分支定界算法, 可变形部件模型算法, 级联检测算法, 梯度直方图特征, 特征金字塔, 隐变量支持向量机, 行人检测

CLC Number:

CHAI Enhui, ZHI Min. Pedestrian detection based on deformable part model with branch and bound[J]. Journal of Computer Applications, 2017, 37(7): 2003-2007.

柴恩惠, 智敏. 融合分支定界的可变形部件模型的行人检测[J]. 计算机应用, 2017, 37(7): 2003-2007.

References

[1] FELZENSZWALB P F, MCALLESTER D, RAMANAN D. A discriminatively trained, multiscale, deformable part model[C]//Proceedings of 2008 IEEE Computer Vision and Pattern Recognition Workshops. Washington, DC:IEEE Computer Society, 2008:1-8.
[2] 刘琦,黄咨,胡福乔,等.基于GPU的卷积检测模型加速[J].计算机应用与软件,2016,33(5):226-230.(LIU Q, HUANG Z, HU F Q, et al. Convolution-based detection models acceleration based on GPU[J]. Computer Applications and Software, 2016, 33(5):226-230.)
[3] PEDERSOLI M, VEDALDI A, GONZAIEZ J, et al. A coarse-to-fine approach for fast deformable object detection[J]. Pattern Recognition, 2015, 48(5):1844-1853.
[4] SONG H, GIRSHICK R, ZICKLER S, et al. Generalized sparselet models for real-time multiclass object recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(5):1001-1012.
[5] 甘鹏坤,陶凌,龙伟.基于可变形部件模型及稀疏特征的行人检测[J].深圳大学学报(理工版),2015,32(6):563-570.(GAN P K, TAO L, LONG W. Cascade pedestrian detection based on the deformable part models and histograms of sparse codes features[J]. Journal of Shenzhen University (Science and Engineering), 2015, 32(6):563-570.)
[6] QIANG Z, SHAI A, MEI C Y, et al. Fast human detection using a cascade of histograms of oriented gradients[C]//Proceedings of 2006 IEEE Conference on Computer Vision and Pattern Recognition Workshops. Washington, DC:IEEE Computer Society, 2006:1491-1498.
[7] 张国云,郭龙源,吴健辉,等.计算机视觉与图像识别[M].北京:科学出版社,2012:165.(ZHANG G Y, GUO L Y, WU J H, et al. Computer Vision and Image Recognition[M]. Beijing:Science Press, 2012:165.)
[8] 田娟,郑郁正.模板匹配技术在图像识别中的应用[J].传感器与微系统,2008,27(1):112-114.(TIAN J, ZHENG Y Z. Application of template matching technique in image recognition[J]. Transducer and Microsystem Technologies, 2008, 27(1):112-114.)
[9] 王宁波.基于RGB-D的行人检测[D].杭州:浙江大学,2013:5-7.(WANG N B. Pedestrian detection based on RGB-D[D]. Hangzhou:Zhejiang University, 2013:5-7.)
[10] ROBERT E S, YORAM S. Improved boosting algorithms using confidence-rated prediction[J]. Machine Learning, 1999, 37(3):297-336.
[11] 程和生,胡幸福.基于HOG和SVM的人体检测技术在静态图像中的研究[J].仪器仪表用户,2012,19(5):20-23.(CHENG H S, HU X F. The study on the human detection technology in the static image based on HOG and SVM[J]. Instrumentation Customer, 2012, 19(5):20-23.)
[12] DALAL N, TRIGGS B. Histograms of oriented gradients for human detection[C]//Proceedings of the 2005 Computer Vision and Pattern Recognition Workshops. Washington, DC:IEEE Computer Society, 2005:886-893.
[13] FELZENSZWALB P F, GIRSHICK R B, MCALLESTER D. et al. Object detection with discriminatively trained part based models[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(9):1627-1645.
[14] FELZENSZWALB P F, HUTTENLOCHER D P. Distance transforms of sampled functions[J]. Theory of Computing, 2012, 8:415-428.
[15] FELZENSZWALB P F, HUTTENLOCHER D P. Pictorial struc-tures for object recognition[J]. International Journal of Computer Vision, 2005, 61(1):55-79.
[16] 赵宇.基于动态规划的检测前跟踪算法研究[D].西安:西安电子科技大学,2012:9-12.(ZHAO Y. Research on tracking algorithm based on dynamic programming[D]. Xi'an:Xidian University, 2012:9-12.)
[17] 张闯.基于可变形部件模型的人体级联检测[D].大连:大连海事大学,2014:33-34.(ZHANG C. Human cascade detection based on deformable component model[D]. Dalian:Dalian Maritime University, 2014:33-34.)
[18] GIRSHICK R B. From rigid templates to grammars:object detection with structured models[D]. Chicago, IL:The University of Chicago, 2012:1-129.

Pedestrian detection based on deformable part model with branch and bound

融合分支定界的可变形部件模型的行人检测

PDF

Knowledge

Abstract

Cite this article

share this article

References

Related Articles 7

Recommended Articles

Metrics

[1]	ZHANG Yangshuo, MIAO Zhuang, WANG Jiabao, LI Yang. Pedestrian detection method based on Movidius neural computing stick [J]. Journal of Computer Applications, 2019, 39(8): 2230-2234.
[2]	HAN Jianghong, YUAN Jiaxuan, WEI Xing, LU Yang. Pedestrian visual positioning algorithm for underground roadway based on deep learning [J]. Journal of Computer Applications, 2019, 39(3): 688-694.
[3]	ZHU Fan, WANG Hongyuan, ZHANG Ji. Fine-grained pedestrian detection algorithm based on improved Mask R-CNN [J]. Journal of Computer Applications, 2019, 39(11): 3210-3215.
[4]	CHEN Guangxi, WANG Jiaxin, HUANG Yong, ZHAN Yijun, ZHAN Baoying. Pedestrian detection method based on cascade networks [J]. Journal of Computer Applications, 2019, 39(1): 186-191.
[5]	LI Lin, ZHANG Tao. Zenithal pedestrian detection algorithm based on improved aggregate channel features and gray-level co-occurrence matrix [J]. Journal of Computer Applications, 2018, 38(12): 3367-3371.
[6]	XU Chao, YAN Shengye. Improved pedestrian detection method based on convolutional neural network [J]. Journal of Computer Applications, 2017, 37(6): 1708-1715.
[7]	MA Qiang, WANG Wenwei. Pedestrian texture extraction by fusing significant factor [J]. Journal of Computer Applications, 2015, 35(11): 3293-3296.