深圳大学学报理工版

针对方向梯度直方图算法无法处理模糊边界且忽略了物体内平滑的特征区域的问题,提出一种基于稀疏编码的可变形部件模型算法.通过稀疏学习得到稀疏编码直方图特征算子的图像特征,利用弱标签隐藏变量结构化支持向量机学习算法对特征进行训练得到部件模型,再结合级联检测算法对人体目标进行识别检测.实验结果显示,混合模型结合级联方法的检测耗时约是混合模型和语义模型平均检测耗时的1/4,与目前其他已有算法比较,所提方法更加鲁棒和具有识别力.

We propose a new sparse encoding based deformable part modelling method to overcome the defect of histogram of orientation gradients algorithm that can not detect fuzzy boundary and smooth feature region inside an object. By using sparse learning, we obtain the image feature operator based on histograms of sparse codes. We use weak label latent variable structured support vector machine to train the feature to derive part model, which is then combined with cascade algorithm to detect human body targets. Experimental results show that the detection time of hybrid model based on cascade method is about a quater that of the hybrid model alone and semantic model. The proposed method has better robustness and recognition ability.

引言
1 图像特征的提取
2 可变形部件模型建模
3 特征训练
4 实验数据分析与处理
5 结语

图1 稀疏编码直方图特征<br/>Fig.1 Histograms of sparse codes features

图1 稀疏编码直方图特征
Fig.1 Histograms of sparse codes features

图2 混合模型中的根滤波器和部件滤波器模型的训练结果<br/>Fig.2 Training results of root and part filters in the mixed model

图2 混合模型中的根滤波器和部件滤波器模型的训练结果
Fig.2 Training results of root and part filters in the mixed model

图3 语义模型中的根滤波器模型和部件滤波器模型的训练结果<br/>Fig.3 Training results of root and part filters in the semantic model

图3 语义模型中的根滤波器模型和部件滤波器模型的训练结果
Fig.3 Training results of root and part filters in the semantic model

图4 单个目标的检测结果<br/>Fig.4 Detection results of a single target

图4 单个目标的检测结果
Fig.4 Detection results of a single target

图5 混合模型与行人语义模型检测结果<br/>Fig.5 Test results of the mixed model and pedestrian semantic model

图5 混合模型与行人语义模型检测结果
Fig.5 Test results of the mixed model and pedestrian semantic model

图6 遮挡目标的检测结果<br/>Fig.6 Detection results of occluded targets

图6 遮挡目标的检测结果
Fig.6 Detection results of occluded targets

图7 检测结果PR曲线<br/>Fig.7 Test results PR curve

图7 检测结果PR曲线
Fig.7 Test results PR curve

表1 级联检测与一般检测耗时比较<br/>Table 1 Comparison of time consume between cascade detection and general detections

表1 级联检测与一般检测耗时比较
Table 1 Comparison of time consume between cascade detection and general detections

[1] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]// International Conference on Computer Vision and Pattern Recognition.[S.l.]: IEEE, 2005:886-893.
[2] Lin Zhe, Davis L S, Doermann D S, et al. Hierarchical part-template matching for human detection and segmentation[C]// IEEE 11th International Conference on Computer Vision. Rio de Janeiro(Brazil): IEEE, 2007: 1-8.
[3] Ren Xiaofeng, Ramanan D. Histograms of sparse codes for object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Portland(USA): IEEE, 2013: 3246-3253.
[4] Elad M, Aharon M. Image denoising via sparse and redundant representations over learned dictionaries[J]. IEEE Transactions on Image Processing, 2006, 15(12): 3736-3745.
[5] Fergus R, Perona P, Zisserman A. Object class recognition by unsupervised scale-invariant learning[C]// Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Madison(USA): IEEE, 2003, 2: II-264-II-271.
[6] Weber M, Welling M, Perona P. Towards automatic discovery of object categories[C]// Proceedings of Computer Vision and Pattern Recognition. Hilton Head Island(USA): IEEE: 101-108.
[7] Tsochantaridis I, Joachims T, Hofmann T, et al. Large margin methods for structured and interdependent output variables[J]. The Journal of Machine Learning Research, 2005,6: 1453-1484.
[8] Lecun Y, Chopra S, Hadsell R, et al. A tutorial on energy-based learning[J]. Predicting Structured Data, 2006.
[9] Zhang Chuang. Human cascade detection based on deformable component model[D]. Dalian: Dalian Maritime University, 2014.(in Chinese)
[10] An Ping. The construction of cascade detector based on linear SVM and its application in target detection[D]. Changsha: National Defense Science and Technology University, 2007.(in Chinese)
[11] Li Tongzhi, Ding Xiaoqing, Wang Shengjin. The human detection method based on cascade[J]. SVM Chinese Journal of Graphics, 2008(3): 566-570.(in Chinese)
[12] Yin Xuecong. Research on face detection method based on deformable component model[D].Xi'an:Xi'an Electronic and Science University, 2012.(in Chinese)
[13] Guo Jie, Zhang Honggang, Chen Daiwu, et al. Object detection algorithm based on deformable part models[C]// Proceedings of the 4th IEEE International Conference on Network Infrastructure and Digital Content. Beijing: IEEE, 2014: 90-94.
[14] Everingham M, Van Gool L, Williams C K I, et al. The pascal visual object classes(VOC)challenge[J]. International Journal of Computer Vision, 2010, 88(2): 303-338.
[15] Everingham M, Ali Eslami S M, Van Gool L, et al. The pascal visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 1111(1): 98-136.

备注

引言

1 图像特征的提取

2 可变形部件模型建模

3 特征训练

4 实验数据分析与处理

5 结语

期刊信息

备注