人体姿态特征选择方法的研究与实现

Research and Implementation of Human Body Posture Feature Selection Method

针对人体姿态特征问题采用宽度作为人体姿态的基本特征,用均分法和改进的遗传算法对宽度特征进行选择,采用BP神经网络分类器对不同的特征定义方法进行典型人体姿态识别的对比实验。实验结果表明,采用该特征定义和选择方法,以每秒12帧的处理速度可达到90%以上的识别正确率。

Aiming at selection method of human body posture, the width is used as basic characteristics of a human body posture, sharing law and improved Genetic Algorithm（GA） are used to choose the width of features, and BP neural network classifier and a number of feature definition approach are used to compare experiments of typical human body posture recognition. Experimental results show that using this feature definition and selection methods to 12 frames per second, processing speed can reach more than 90% recognition accuracy rate.