为实现在复杂背景下对人手进行准确、快速的位置跟踪与轮廓跟踪，并针对粒子滤波无法获取人手目标的准确信息与人手深凹区域获取难的问题，提出一种基于肤色自适应梯度矢量流主动轮廓(Gradient Vector Flow Snake, GVF Snake)模型的粒子滤波算法实现人手跟踪.该算法首先对粒子滤波得到的人手区域进行肤色灰度增强，弱化背景梯度信息，然后对该区域利用引入自适应梯度矢量流场和肤色自适应外部引导力的GVF Snake模型，获取人手的真实轮廓以及准确的系统观测和系统状态，减少了粒子退化的可能，完成了更加准确的位置跟踪.实验表明：在复杂、运动背景甚至大范围遮挡的情况下，该改进算法获得了更加准确的人手轮廓，同时对人手跟踪的实时性提升了13%，均方根误差降低了48%.
To implement the hand tracking and contour tracking accurately and quickly in complex background, a novel algorithm for hand tracking with particle filter and skin color adaptive gradient vector flow snake (GVF snake) model is proposed. This algorithm especially applies to the extraction of deep concave region of hand contour, and overcomes the problem that particle filter can not obtain accurate information. Firstly, the hand region obtained by the particle filter is grayscale enhanced by skin color, weakening the background gradient information. And then the GVF snake model, with adaptive gradient vector flow field and adaptive external guidance force improved by skin color, is introduced to extract the real contour and modify the systematic observation and system state of particle filter. In this case, it can reduce the possibility of degradation of particles and realize hand tracking more accurately. The experimental results show that the proposed algorithm extracts more accurate contour of hand and improves real-time performance of human tracking by 13% and the root mean square error is reduced by 48%, under the conditions of complexity, moving background and even a wide range of occlusion.
［1］WANG X, ZHANG X, DAI G. Tracking of deformable human hand in real time as continuous input for gesture-based interaction［C］∥Proceedings of Inter-national Conference on Intelligent User Interfaces. Honolulu, Hawaii, USA: DBLP, 2007: 235-242.
［2］刘法旺, 丁刚毅, 李善青, 等. 基于ICONDENSATION算法的人手跟踪与手势识别算法［J］. 北京理工大学学报, 2007, 27(12): 1069-1072.
LIU Fawang, DING Gangyi, LI Shanqing, et al. ICONDENSATION-based hand tracking and gesture recognition with color and depth cues［J］. Transactions of Beijing Institute of Technology, 2007, 27(12): 1069-1072.
［3］LIM K M, TAN A W C, TAN S C. A feature cova-riance matrix with serial particle filter for isolated sign language recognition［J］. Expert Systems with Applications: An International Journal, 2016, 54(C): 208-218.
［4］季玉龙, 杨光, 戈文一. 基于粒子滤波的目标主动轮廓跟踪算法［J］. 电子科技大学学报, 2015, 41(1): 91-96.
JI Yulong, YANG Guang, GE Wenyi. A novel particle filter based object active contour tracking method［J］. Journal of the University of Electronic Science and Technology of China, 2015, 44(1): 91-96.
［5］ZHAO S, SONG X, TAN W, et al. A novel approach to hand gesture contour detection based on GVF Snake model and skin color elliptical model ［C］∥International Conference on Computer Application and System Modeling. Kuala Lumpur, Malaysia: IEEE, 2010: V5-381-V5-384.
［6］李晓旭, 戴彬, 曹洁. 基于多特征融合的改进UPF目标跟踪算法［J］. 上海交通大学学报, 2014, 48(10): 1473-1478.
LI Xiaoxu, DAI Bin, CAO Jie. An improved UPF object tracking algorithm based on multi-feature fusion［J］. Journal of Shanghai Jiao Tong University, 2014, 48(10): 1473-1478.
［7］GONZALES C, DUBUISSON S. Combinatorial re-sampling particle filter: An effective and efficient method for articulated object tracking［J］. International Journal of Computer Vision, 2015, 112(3): 255-284..
［8］张铁, 马琼雄. 人机交互中的人体目标跟踪算法［J］. 上海交通大学学报, 2015, 49(8): 1213-1219.
ZHANG Tie, MA Qiongxiong. Human object tracking algorithm for human-robot interaction［J］. Journal of Shanghai Jiao Tong University, 2015, 49(8): 1213-1219.
［9］ONGKITTIKUL S, WORRALL S, KONDOZ A. Two hand tracking using colour statistical model with the k-means embedded particle filter for hand gesture recognition［C］∥Proceedings of Computer Information Systems and Industrial Management Applications. IEEE Computer Society. Washington, DC, USA: IEEE Computer Society, 2008: 201-206.
［10］HESS R, FERN A. Discriminatively trained particle filters for complex multi-object tracking［C］∥IEEE Conference on Computer Vision & Pattern Recognition. Miami, Florida, USA: IEEE Computer Society, 2009: 240-247.
［11］TANG J. A multi-direction GVF Snake for the segmentation of skin cancer images［J］. Pattern Recognition, 2009, 42(6): 1172-1179.
［12］ZHAI P, SHI C. Normal and tangent components normalization based GVF Snake for image segmentation［C］∥International Conference on Digital Image Processing. Chengu, Chin: International Society for Optics and Photonics, 2016: 1003318-1003319.
［13］XU C, YEZZI A J, PRINCE J L. On the relationship between parametric and geometric active contours［C］∥Asilomar Conference on Signals, Systems and Computers. Pacific Grove, CA, USA: Circuits, Systems and Computers, 2000: 483-489.
［14］XIANG Y, CHUNG A C S, YE J. An active contour model for image segmentation based on elastic interaction［J］. Journal of Computational Physics, 2006, 219(1): 455-476.
［15］SMISTAD E, ELSTER A C, LINDSETH F. Real-time gradient vector flow on GPUs using OpenCL［J］. Journal of Real-Time Image Processing, 2015, 10(1): 67-74.
