基于矢状面和神经网络的三维人体骨架提取  

3D Human Skeleton Extraction Based on Sagittal Plane and Neural Network

作　　者：黄新[1] 郝矿荣[1,2] 丁永生[1,2] 

机构地区：[1]东华大学信息科学与技术学院,上海201620 [2]数字化纺织服装技术教育部工程研究中心,上海201620

出　　处：《计算机工程》2012年第4期14-16,共3页Computer Engineering

基　　金：国家自然科学基金资助重点项目(61134009);国家自然科学基金资助项目(60975059;60775052);国家ITER计划国内配套研究基金资助项目(2010GB108004);教育部高等学校博士学科点专项科研基金资助项目(20090075110002);上海市优秀学术带头人计划基金资助项目(11XD1400100);上海市科学技术委员会重点基础研究基金资助项目(10JC1400200;09JC1400900);上海市科学技术委员会技术标准专项基金资助项目(10DZ0506500)

摘　　要：不同姿态的人体模型易对骨架提取算法产生干扰。为此,提出一种新的骨架提取算法。该算法通过将人体模型矢状面深度信息和改进Hopfield神经网络相结合的方式,引入一种网络输入输出函数,对传统的人体骨架提取算法进行改进,使网络收敛速度明显加快。通过特征点的深度信息决定点对差异的方式,使网络成功地避免局部极小点,同时减少网络的运行时间。实验结果表明,该算法在定位骨架特征点处的误差明显小于传统算法,且缩短了算法的运行时间。该算法对人体骨架提取的效果更好。In allusion to the problem of interference to the skeleton extraction of different human posture, a new algorithm of skeleton extraction is presented. Through the combination of the depth information based on sagittal plane of 3D human model with improved Hopfield neural network, the rate of convergence speeds up by using a new input-output function of network to improve traditional human skeleton extraction algorithm. The network gets away from local minimum successfully and decreases the running time of network which is decided to depth information of feature points. Experimental result shows that the displacement on skeleton feature points using new algorithm is obvious less than that of traditional algorithm. In addition, the computation time is decreased. Therefore, the new algorithm has better effect on human skeleton extraction.

关 键 词：骨架提取 人体模型 矢状面 深度信息 HOPFIELD网络 特征点匹配 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]