基于关节点定位的重点人群步态分类方法研究

Research on Gait Classification Method of Key Population Based on Joint Point Location

基于模型的主流步态识别方法通常过于关注下肢运动而忽视上肢信息。为解决该方法中上肢信息利用率不足的问题,文章通过改进的ZS细化算法提取到更完整的上肢骨架,同时提出一种基于关节点定位的步态特征提取方法应用于重点人群步态分类。将细化所得人体骨架与行人轮廓图相结合,定位全身10个关节点坐标并建立人体模型,利用关节角度的时序特征进行分类效果验证,在中科院CASIA-B数据集上进行实验。实验结果表明:该文所提取的步态特征有较高的识别性能,平均步态识别率为77.35%,同时使用上下肢进行步态分类的平均召回率为84%,与仅采用下肢信息相比提高38%。

The mainstream gait recognition methods based on model usually pay too much attention to the movement of lower limbs and ignore the information of upper limbs.In order to solve the problem of insufficient utilization of upper limb information in this method,a more complete upper limb skeleton is extracted through the improved ZS thinning algorithm,and a gait feature extraction method based on joint point localization is proposed to apply to the gait classification of key populations.The human skeleton obtained by thinning algorithm is combined with the pedestrian contour map to locate the coordinates of 10 joint points of the whole body.Then the human model could be established.The classification effect is verified by using the time series feature of joint angle,and the experiment is carried out on CASIA-B dataset of Chinese Academy of Sciences.The experimental results show that the gait features extracted in this paper have high recognition performance.The average gait recognition rate is 77.35%,and the average recall rate of gait classification using upper and lower limbs is 84%,which is 38%higher than that using only lower limb information.