融合密集卷積與空間轉(zhuǎn)換網(wǎng)絡(luò)的手勢識別方法
doi: 10.11999/JEIT170627 cstr: 32379.14.JEIT170627
基金項目:
國家自然科學(xué)基金(61203245)，河北省自然科學(xué)基金(F2012202027)
Gesture Recognition Method Combining Dense Convolutional with Spatial Transformer Networks
Funds:
The National Natural Science Foundation of China (61203245), The Natural Science Foundation of Hebei Province (F2012202027)
-
摘要: 手勢識別作為人機交互的方式之一，在人工智能日益發(fā)展的今天備受矚目。針對手勢旋轉(zhuǎn)、平移、縮放等形變導(dǎo)致識別率偏低的問題，該文基于密集卷積網(wǎng)絡(luò)(Densenet)與空間轉(zhuǎn)換網(wǎng)絡(luò)(STN)提出了一種新型的網(wǎng)絡(luò)結(jié)構(gòu)Densenet_V2，先利用空間轉(zhuǎn)換網(wǎng)絡(luò)對輸入的樣本和特征圖進行空間變換和對齊，再利用密集卷積網(wǎng)絡(luò)自動提取手勢的有效特征，最后通過線性分類器對手勢進行分類。為防止網(wǎng)絡(luò)模型對樣本數(shù)據(jù)集過度擬合，對網(wǎng)絡(luò)進行訓(xùn)練時在損失函數(shù)中加入L2正則項以實現(xiàn)權(quán)重衰減。在Marcel手勢庫上進行多次實驗。實驗結(jié)果表明，Densenet_V2可以提高對靜態(tài)形變手勢的識別率。
-
關(guān)鍵詞:
- 手勢識別 /
- 形變 /
- 密集卷積網(wǎng)絡(luò) /
- 空間轉(zhuǎn)換網(wǎng)絡(luò) /
- L2正則項
Abstract: As an important milestone for the development of the artificial intelligence, gesture recognition enables the human-computer interaction and has received significantly growing research interest nowadays. However, the current technology for the gesture recognition has the low quality in the gesture rotation, translation and scaling. To solve the problem, a novel network structure named Densenet_V2 is proposed, and it is based on Dense Convolutional Networks (Densenet) and Spatial Transformer Networks (STN). Firstly, the input samples and feature maps are spatially transformed and aligned with the STN. Then the effective features of gestures are automatically extracted by using the Densenet. Finally, the linear classifier is adopted to classify the gestures. To prevent the network model from over-fitting the sample data set, the L2 regular term is involved into the loss function to achieve the weight decay when training the network. Experiments on the Marcel gesture database show that Densenet_V2 can improve the recognition rate of static deformation gestures. -
PIYUSH K, SIDDHARTH S R, and Anupam A. Hand data glove: A new generation real-time mouse for human- computer interaction[C]. International Conference on Recent Advances in Information Technology (RAIT), Dhanbad, Jharkand, India, 2012: 750-755. doi: 10.1109/RAIT.2012. 6194548. WEI W and JING P. Hand segmentation using skin color and background information[C]. International Conference on Machine Learning and Cybernetics, Xi,an, China, 2012: 1487-1492. doi: 10.1109/ICMLC.2012.6359584. 阮曉鋼, 林佳, 于乃功, 等. 基于多線索的運動手部分割方法[J]. 電子與信息學(xué)報, 2017, 39(5): 1088-1095. doi: 10.11999/ JEIT160730. RUAN Xiaogang, LIN Jia, YU Naigong, et al. Moving hand segmentation based on multi-cues[J]. Journal of Electronics Information Technology, 2017, 39(5): 1088-1095. doi: 10. 11999/JEIT160730. LIU Y, YIN Y, and ZHANG S. Hand gesture recognition based on HU moments in interaction of virtual reality[C]. International Conference on Intelligent Human-Machine Systems and Cybernetics, Nanchang, China, 2012: 145-148. doi: 10.1109/IHMSC.2012.42. DARDAS N H and GEORGANAS N D. Real-time hand gesture detection and recognition using bag-of-features and support vector machine techniques[J]. IEEE Transactions on Instrumentation Measurement, 2011, 60(11): 3592-3607. doi: 10.1109/TIM.2011.2161140. 楊學(xué)文, 馮志全, 黃忠柱, 等. 結(jié)合手勢主方向和類- Hausdorff距離的手勢識別[J]. 計算機輔助設(shè)計與圖形學(xué)學(xué)報, 2016, 28(1): 75-81. doi: 10.3969/j.issn.1003-9775.2016.01.010. YANG Xuewen, FENG Zhiquan, HUANG Zhongzhu, et al. Gesture recognition based on combining main direction of gesture and Hausdorff-like distance[J]. Journal of Computer- Aided Design Computer Graphics, 2016, 28(1): 75-81. doi: 10.3969/j.issn.1003-9775.2016.01.010. 劉淑萍, 劉羽, 於俊, 等. 結(jié)合手指檢測和HOG特征的分層靜態(tài)手勢識別[J]. 中國圖象圖形學(xué)報, 2015, 20(6): 781-788. doi: 10.11834/jig.20150607. LIU Shuping, LIU Yu, YU Jun, et al. Hierarchical static hand gesture recognition by combining finger detection and HOG features[J]. Journal of Image and Graphics, 2015, 20(6): 781-788. doi: 10.11834/jig.20150607. LIN H I, HSU M H, and CHEN W K. Human hand gesture recognition using a convolution neural network[C]. IEEE International Conference on Automation Science and Engineering, Taipei, China, 2014: 1038-1043. doi: 10.1109/ CoASE.2014.6899454. 杜堃, 譚臺哲. 復(fù)雜環(huán)境下通用的手勢識別方法[J]. 計算機應(yīng)用, 2016, 36(7): 1965-1970. doi: 10.11772/j.issn.1001-9081. 2016.07.1965. DU Kun and TAN Taizhe. General method for gesture recognition in complex environment[J]. Journal of Computer Applications, 2016, 36(7): 1965-1970. doi: 10.11772/j.issn. 1001-9081.2016.07.1965. PYO J, JI S, and YOU S. Depth-based hand gesture recognition using convolutional neural networks[C]. International Conference on Ubiquitous Robots and Ambient Intelligence, Xi,an, China, 2016: 225-227. doi: 10.1109/URAI. 2016.7625742. LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. doi: 10.1109/5.726791. JADERBERG M, SIMONYAN K, ZISSERMAN A, et al. Spatial transformer networks[OL]. https://arxiv.org/abs/ 1506.02025v3,2015. LECUN Y, BENGIO Y, and HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444. doi: 10.1038/nature14539. GOODFELLOW I, BENGIO Y, and COURVILLE A. Deep Learning[M]. Massachusetts, USA: MIT Press, 2016: 231-234. -
計量
- 文章訪問數(shù): 1365
- HTML全文瀏覽量: 187
- PDF下載量: 221
- 被引次數(shù): 0
下載:
