| 引用本文: | 李昱,季文彬,戴士杰.全身步态模型的视触融合步态识别算法[J].哈尔滨工业大学学报,2022,54(1):88.DOI:10.11918/202012088 |
| LI Yu,JI Wenbin,DAI Shijie.Visual-tactile fusion gait recognition based on full-body gait model[J].Journal of Harbin Institute of Technology,2022,54(1):88.DOI:10.11918/202012088 |
|
| 摘要: |
| 为减少背包负重、衣着和环境等因素对步态识别率的影响,提出一种融合视觉和触觉特征的全身步态模型。首先,以支撑脚为起点,根据运动传递过程,建立身体各个部分质量与地面支持力的动力学关系,并且通过加速度引入视觉特征；然后,对模型进行参数分离,得到代表不同步态运动特征的特征矩阵,利用Kinect和步道式足底压力仪获得的视觉图像序列和足底压力图像提取视觉和触觉特征,建立包含正常、背包负重和穿大衣3种步态运动状态下的数据库；最后,选择支持向量机中的多分类方法完成步态识别,在识别过程中通过K-CV法对分类器参数进行了寻优。实验结果表明:足底压力分区方式增加了特征识别点,提高了模型识别率；在正常步态运动条件下模型平均识别率为97.31%,在背包和穿大衣的情况下模型识别性能下降比较少。融合视觉和触觉特征建立包含上肢摆动的全身步态模型可以有效提高模型在复杂步态运动条件下的鲁棒性和步态识别准确率。 |
| 关键词: 步态识别 全身步态运动模型 视触融合 特征提取 支持向量机 |
| DOI：10.11918/202012088 |
| 分类号:TP273 |
| 文献标识码:A |
| 基金项目:国家重点研发计划(2019YFB1311104) |
|
| Visual-tactile fusion gait recognition based on full-body gait model |
|
LI Yu1,2,JI Wenbin1,2,DAI Shijie1,2
|
|
(1. State Key Laboratory of Reliability and Intelligence of Electrical Equipment(Hebei University of Technology),Tianjin 300130, China; 2. School of Mechanical Engineering,Hebei University of Technology,Tianjin 300000,China)
|
| Abstract: |
| To reduce the influence of factors such as backpack load, clothing and environment on gait recognition rate, a full-body gait model fusing visual and tactile features was proposed. The model first took the support foot as the starting point, established the kinetic relationship between the mass of each body part and the ground support force according to the motion transfer process, and introduced visual features through acceleration. Then the model was parameter separated to obtain feature matrices representing different gait motion features, the visual and tactile features were extracted using visual image sequences and plantar pressure images obtained from Kinect and walkway-type plantar pressure meter. A database containing the three gait motion states of normal, backpack loaded and overcoat wearing was established. Finally the multi-classification method in support vector machine was selected to complete the gait recognition, and the classifier parameters were optimized by the K-CV method in the recognition process. The experimental results showed that the model recognition rate was improved by increasing the number of feature recognition points by means of plantar pressure partitioning. The average recognition rate of the model under normal gait motion conditions was 97.31%, and the recognition performance of the model decreased less in the case of backpack and wearing a coat. Fusion of visual and tactile features to build a full-body model including upper limb swing could effectively improve the robustness of the model under complex gait motion conditions and increase the gait recognition accuracy. |
| Key words: gait recognition the full-body gait model fusion of vision and tactile feature extraction support vector machine |
