基于空间金字塔池化特征的日常工具分类识别
|
摘要
面向人机共融环境下机器智能对工具认知的需要,为提高家庭服务机器人的工具功能用途认知能力,设计一种基于深度几何特征空间金字塔池化的工具功用性建模与分类方法.离线训练阶段,考虑到各类工具在几何形态上的差异对工具自身更具表征性,首先,在各工具的深度图上提取多类几何特征,并融合形成工具特征图;然后,在工具特征图上提取多尺度特征块,并基于cciPCA的方法建立空间池化金字塔,从而构建最终的工具特征向量;最后,在高层语义空间上,利用SVM分类器训练工具分类识别模型.在线检测阶段,利用离线训练的工具分类模型对空间池化的样本进行分类测试.实验结果表明,所提方法能够实现家庭服务机器人对家庭日常工具的认知及分类识别,部分工具的识别精度可达97%及以上.
In view of the need of machine intelligence for tool recognition in man-machine communion environment, in order to improve the ability for home service robot to recognize affordances of tools, a modeling and classification method based on depth geometry feature spatial pyramid pooling is designed. In the off-line training phase, considering that the differences in the geometric shapes of various tools are more indicative for the tools themselves, firstly, the multi-types of geometric features are extracted from each tool depth map and fused to form the tool feature map. Then, the multi-scale feature patches are extracted and the spatial pooled pyramid is built based on cciPCA to construct the final tool feature vector. Finally, in the high-level semantic space, the tools classification model is trained based on the SVM classifier. In the online testing phase, the samples of spatial pooling are tested by a tool model trained offline. Experimental results show that the proposed method realizes recognition and classification of daily tools for home service robots, and the recognition accuracy of some tools reaches 97% and above.
In view of the need of machine intelligence for tool recognition in man-machine communion environment, in order to improve the ability for home service robot to recognize affordances of tools, a modeling and classification method based on depth geometry feature spatial pyramid pooling is designed. In the off-line training phase, considering that the differences in the geometric shapes of various tools are more indicative for the tools themselves, firstly, the multi-types of geometric features are extracted from each tool depth map and fused to form the tool feature map. Then, the multi-scale feature patches are extracted and the spatial pooled pyramid is built based on cciPCA to construct the final tool feature vector. Finally, in the high-level semantic space, the tools classification model is trained based on the SVM classifier. In the online testing phase, the samples of spatial pooling are tested by a tool model trained offline. Experimental results show that the proposed method realizes recognition and classification of daily tools for home service robots, and the recognition accuracy of some tools reaches 97% and above.
引文
[1]Zou Fuhao,Wang Yunfei,Yang Yang,et al.Supervised feature learning via l2-norm regularized logistic regression for 3D object recognition[J].Neurocomputing2015,151(2):603-611.
[2]Kjellstrom H,Romero J,Kragi D.Visual object-action recognition:Inferring object affordances from human demonstration[J].Computer Vision and Image Understanding,2011,115(1):81-90.
[3]Zhu Y X,Zhao Y B,Zhu S C.Understanding tools:Task-oriented object modeling,learning and recognition[C].Proc of the IEEE Conf on Computer Vision and Pattern Recognition.Boston:IEEE,2015:2855-2864.
[4]Mahmudul H,Dharmaratne A.Attribute based affordance detection from human-object interaction images[C].Proc of the Image and Video Technology-PSIVT Workshops New Zealand:Springer Verlag,2015:220-232.
[5]Edsinger A,Kemp C C.Manipulation in human environments[C].IEEE-RAS Int Conf on Humanoid Robots.Genoa:IEEE,2006:13-19.
[6]Tanis M,Vadim T,Giorgio M.Multi-model approach based on 3D functional features for tool affordance learning in robotics[C].IEEE-RAS Int Conf on Humanoid Robots.Seoul:IEEE,2015:482-489.
[7]Lenz I,Lee H,Saxena A.Deep learning for detecting robotic grasps[J].Int J of Robotics Research,201534(4/5):705-724.
[8]Redmon J,Angelova A.Real-time grasp detection using convolutional neural networks[C].IEEE Int Conf on Robotics and Automation.Seattle:IEEE,2015:26-30.
[9]Nguyen A,Kanoulas D,Caldwell D G,et al Detecting object affordances with convolutional neural networks[C].Proc of IEEE/RSG Int Conf on Intelligent Robots and Systems.IEEE,2016:2765-2770.
[10]吴培良,付卫兴,孔令富.一种基于结构随机森林的家庭日常工具部件功用性快速检测算法[J].光学学报2017(2):155-165.(Wu P L,Fu W X,Kong L F.A fast algorithm for affordance detection of household tool parts based on structured random forest[J].Acta Optica Sinica,2017(2):155-165.)
[11]Grauman K,Darrell T.The pyramid match kernel:Discriminative classification with sets of image features[C].The 10th IEEE Int Conf on Computer Vision Beijing:IEEE,2005:1458-1465.
[12]Lazebnik S,Schmid C,Ponce J.Beyond bags of features:Spatial pyramid matching for recognizing natural scene categories[C].IEEE Computer Society Conf on Computer Vision and Pattern Recognition.New York:IEEE,2006:2169-2178.
[13]Yang J,Yu K,Gong Y,et al.Linear spatial pyramid matching using sparse coding for image classification[C]IEEE Computer Society Conf on Computer Vision and Pattern Recognition.Miami:IEEE,2009:1794-1801.
[14]Shen F,Shen C,Zhou X,et al.Face image classification by pooling raw features[J].Pattern Recognition,201654(C):94-103.
[15]He K,Zhang X,Ren S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J]IEEE Trans on Pattern Analysis&Machine Intelligence2014,37(9):1904-1916.
[16]Han X,Zhong Y,Cao L,et al.Pre-trained alexNet architecture with pyramid pooling and supervision for high spatial resolution remote sensing image scene classification[J].Remote Sensing,2017,9(8):848.
[17]Myers A,Teo C L,Fermüller C,et al.Affordance detection of tool parts from geometric features[C].IEEEInt Conf on Robotics and Automation.Seattle:IEEE2015:1374-1381.
[18]Koenderink J J,Van D A J.Surface shape and curvature scales[J].Image and Vision Computing,1992,10(8):557-564.
[19]Coates A,Ng A.The importance of encoding versus training with sparse coding and vector quantization[C]Int Conf on Machine Learning.Bellevue:Association for Computing Machinery,2011:921-928.
[2]Kjellstrom H,Romero J,Kragi D.Visual object-action recognition:Inferring object affordances from human demonstration[J].Computer Vision and Image Understanding,2011,115(1):81-90.
[3]Zhu Y X,Zhao Y B,Zhu S C.Understanding tools:Task-oriented object modeling,learning and recognition[C].Proc of the IEEE Conf on Computer Vision and Pattern Recognition.Boston:IEEE,2015:2855-2864.
[4]Mahmudul H,Dharmaratne A.Attribute based affordance detection from human-object interaction images[C].Proc of the Image and Video Technology-PSIVT Workshops New Zealand:Springer Verlag,2015:220-232.
[5]Edsinger A,Kemp C C.Manipulation in human environments[C].IEEE-RAS Int Conf on Humanoid Robots.Genoa:IEEE,2006:13-19.
[6]Tanis M,Vadim T,Giorgio M.Multi-model approach based on 3D functional features for tool affordance learning in robotics[C].IEEE-RAS Int Conf on Humanoid Robots.Seoul:IEEE,2015:482-489.
[7]Lenz I,Lee H,Saxena A.Deep learning for detecting robotic grasps[J].Int J of Robotics Research,201534(4/5):705-724.
[8]Redmon J,Angelova A.Real-time grasp detection using convolutional neural networks[C].IEEE Int Conf on Robotics and Automation.Seattle:IEEE,2015:26-30.
[9]Nguyen A,Kanoulas D,Caldwell D G,et al Detecting object affordances with convolutional neural networks[C].Proc of IEEE/RSG Int Conf on Intelligent Robots and Systems.IEEE,2016:2765-2770.
[10]吴培良,付卫兴,孔令富.一种基于结构随机森林的家庭日常工具部件功用性快速检测算法[J].光学学报2017(2):155-165.(Wu P L,Fu W X,Kong L F.A fast algorithm for affordance detection of household tool parts based on structured random forest[J].Acta Optica Sinica,2017(2):155-165.)
[11]Grauman K,Darrell T.The pyramid match kernel:Discriminative classification with sets of image features[C].The 10th IEEE Int Conf on Computer Vision Beijing:IEEE,2005:1458-1465.
[12]Lazebnik S,Schmid C,Ponce J.Beyond bags of features:Spatial pyramid matching for recognizing natural scene categories[C].IEEE Computer Society Conf on Computer Vision and Pattern Recognition.New York:IEEE,2006:2169-2178.
[13]Yang J,Yu K,Gong Y,et al.Linear spatial pyramid matching using sparse coding for image classification[C]IEEE Computer Society Conf on Computer Vision and Pattern Recognition.Miami:IEEE,2009:1794-1801.
[14]Shen F,Shen C,Zhou X,et al.Face image classification by pooling raw features[J].Pattern Recognition,201654(C):94-103.
[15]He K,Zhang X,Ren S,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J]IEEE Trans on Pattern Analysis&Machine Intelligence2014,37(9):1904-1916.
[16]Han X,Zhong Y,Cao L,et al.Pre-trained alexNet architecture with pyramid pooling and supervision for high spatial resolution remote sensing image scene classification[J].Remote Sensing,2017,9(8):848.
[17]Myers A,Teo C L,Fermüller C,et al.Affordance detection of tool parts from geometric features[C].IEEEInt Conf on Robotics and Automation.Seattle:IEEE2015:1374-1381.
[18]Koenderink J J,Van D A J.Surface shape and curvature scales[J].Image and Vision Computing,1992,10(8):557-564.
[19]Coates A,Ng A.The importance of encoding versus training with sparse coding and vector quantization[C]Int Conf on Machine Learning.Bellevue:Association for Computing Machinery,2011:921-928.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |