人体运动姿态图像矫正识别仿真研究

Three Dimensional Motion Pattern Recognition Posture Correction

在对姿态图像进行视觉矫正的过程中,人的运动存在较大的随机性,导致三维运动参数无法在一定区域内固定。当运动参数发生变化时,会导致传统的姿态识别在参数对比过程中,姿态特征的类内差异与类间差异过小无法提取的问题。导致姿态矫正精度低、准确性差。提出一种将三维运动形态识别应用于姿态矫正的方法,通过Gabor滤波器及图像的Gabor小波特征,提取运动图像的二维Gabor小波特征,检测正面运动形态,并定位一张正面运动形态和一张运动形态图像中重要的运动形态特征点。利用采集的二维Gabor特征向量与一个常用的3D运动形态数据库重建三维运动形态模型。并通过模板匹配与线性判别分析对其进行处理,获取模型的类内差异与类间差异,实现三维运动形态矫正。依据运动图像在三维空间中的旋转所投影的平面图像,实现运动图像的姿态矫正。仿真结果表明,所提方法具有较高的矫正精度。

A posture correction method is presented based on the pattern recognition of three - dimensional move- ment. Through the Gabor filter and image Gabor small potter character, motion images of 2d Gabor wavelet are ex- tracted, positive movement forms are detected, and the important movement feature points in images of a positive movement form and a movement form are located. Use 2d Gabor feature vector and a common form of 3d motion data- base are used to reconstruct the 3d motion form model. By template matching and linear discriminant analysis for its processing, the model within the class differences and differences between classes is obtained, and the three - dimen- sional movement form correction is implemented. Based on motion image rotation in the three - dimensional space projection plane image, the posture correction of motion image is achieved. Simulation results show that the proposed method has high accuracy.