基于统计模型的遗传粒子滤波器人体运动跟踪

Human motion tracking based on statistical model and genetic particle filter

提出了一种基于统计模型的遗传粒子滤波器人体运动跟踪算法。引入局域二值模式(LBP)算子提取纹理特征,利用颜色直方图与纹理直方图相似度的加权和表示目标相似度,以有效解决自遮挡对跟踪的影响。利用该统计模型精确表示运动人体轮廓,目标形状可由一可变形状参数确定;采用遗传粒子滤波器作为跟踪算法以提高粒子滤波器的鲁棒性和精度。通过预测更新可变形状参数,再利用统计模型中目标形状与形状可变参数的关系得到图像序列各帧中人体轮廓,有效降低了计算量,从而达到快速而准确的跟踪目的。最后用上述方法进行了实验,验证了该方法的实用性和有效性。

A method of human motion tracking was presented, which was based on statistical model and genetic particle filter. The local binary pattern was introduced to extract the texture feature. And to solve the self-occlusion problem, the similarity was expressed by the sum of the weighted color and texture histogram similarity. The human silhouette was exactly presented by a statistical model, and through the variable shape parameter the shape vector of the object could be get. The genetic particle filter was used as the tracking method to improve the robustness and precision. First, the variable shape parameter could be predicted and updated in genetic particle filter, and then through the connection between the variable shape parameter and the shape vector of the object, the human silhouette in the image sequence could be get. This method could effectively reduce the computation cost and realize the purpose that tracks the human motion quickly and exactly. Finally, it also performed some experiments, and proved that the method works effectively.