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.