动态分层二值进化全关节驱动跟踪贫化算法

Tracking Dilution Method of Whole Joint Driving Based on Dynamic Hierarchical Binary Evolution

在人体运动跟踪建模中,需要对样本集的多样性特征进行贫化处理,以提高全关节驱动模式运动状态跟踪的准确性。传统方法采用量子进化和粒子滤波算法进行人体运动跟踪贫化算法实现,算法在全关节多样化样本特征运动模式下,跟踪效果不好。提出一种采用动态分层二值进化处理的改进的量子进化粒子滤波全关节驱动模式跟踪方法,解决多样本特征的贫化问题。进行人体全关节驱动模式动力学分析及人体运动跟踪模型构建,通过动态分层处理技术,获得二值前景图像,求得人体关节的全方位信息特征,通过动态分层二值进化方法,准确地找到各关节位置,构建亮度模型函数,实现贫化处理。实验表明,改进算法能实现对体操运动员运动幅度大的肘、腕、踝部位均得到了准确的跟踪结果,贫化效果较好,运动状态估计精度较高。

In the tracking of human motion modeling, the need for diversity of characteristics of sample set for dilution pro-cess, in order to improve the accuracy of the motion state of the joint driving mode tracking. The traditional methods of quantum evolution and particle filtering algorithm for human motion tracking and dilution algorithm, algorithm diverse sam-ple characteristic movement patterns in the joint, the tracking effect is not good. A dynamic hierarchical value of two im-proved evolutionary process of quantum evolutionary particle filter tracking method for total joint drive mode is proposed, the dilution problem of multi sample characteristics. Human total joint drive mode dynamics analysis and tracking of hu-man motion model, the dynamic slicing technology gets a value of two, the foreground image, a full range of information characteristics obtained by dynamic hierarchical human body joints, two value evolution method, accurately find the joint position, constructing the brightness model function, realize the dilution process. Experiments show that the improved algo-rithm can realize the gymnastics athletes elbow, wrist, high amplitude of ankle positions are accurate tracking results, the dilution effect is good, the motion state estimation accuracy is improved.