基于随机森林算法的服装3D人体特征识别

3D Human Body Feature Recognition Based on Random Forest

提出了一种基于随机森林算法的3D人体特征识别算法.首先,利用bootsrap重抽样从3D人体特征样本中抽取多个样本,并对每个bootsrap样本进行建模,生成一定数量的决策树,在此基础上组合多个决策树的预测,通过投票预测特征点,把投票比例最高的点作为特征点.然后,利用3次B样条对特征点进行拟合得到3D扫描人体轮廓线,并测定人体尺寸数据.最后,将测试结果与标准测量结果进行比较,计算误差值.仿真实验表明,该方法对不同的3D扫描人体模型具有良好的识别效果.

A new method to recognize 3D human body features based on random forests was proposed in this paper. Firstly,resample was used to extract samples from the 3D features of human samples,and ensuring each Boots rap sample to generate a certain number of decision trees,on this basises combining a plurality of decision trees forecast,by a final voteto forecast the feature points,the highest vote scale points of the feature points as the last feature points. Second,three B-spline was used to fit the feature points to get 3D scanning of body contours lines,and measure body size data.Finally,the test results and the results were compared with standard measurement,and calculated the error value. Simulation results show that the simulation have good recognition to various 3D scanhuman body.