欢迎访问《图学学报》 分享到:

图学学报

• 专论:第18届中国虚拟现实大会暨第8届国际虚拟现实与可视化会议(ChinaVR & ICVRV 2018 青岛 ) • 上一篇    下一篇

基于少量关键序列帧的人体姿态识别方法

  

  1. (北方工业大学信息学院,北京 100144)
  • 出版日期:2019-06-30 发布日期:2019-08-02
  • 基金资助:
    国家自然科学基金项目(61503005);北京市自然科学基金项目(4162022);北方工业大学长城学者培养项目(NCUTCC08)

Human Posture Recognition Method Based on Few Key Frames Sequence

  1. (School of Information Science and Technology, North China University of Technology, Beijing 100144, China)
  • Online:2019-06-30 Published:2019-08-02

摘要: 针对传统人体姿态识别数据采集易受环境干扰、难以解决人体运动姿态的相似性和 人体运动执行者的特征差异性等问题,提出一种基于少量关键序列帧的人体姿态识别方法。首先 对原有运动序列进行预选,通过运动轨迹取极值的方法构造初选关键帧序列,再利用帧消减算法 获取最终关键帧序列;然后对不同人体姿态分别建立隐马尔科夫模型,利用 Baum-Welch 算法计 算得到初始概率矩阵、混淆矩阵、状态转移矩阵,获得训练后模型;最后输入待测数据,应用前 向算法,得到对于每个模型的概率,比较并选取最大概率对应的姿态作为识别结果。实验结果表 明,该方法能够有效的选取原始运动序列的关键帧,提高人体姿态识别的准确性。

关键词: 人体姿态识别, 序列帧, 帧消减, 隐马尔科夫模型

Abstract:  This study focuses on the problems that the traditional human posture recognition data acquisition is easily disturbed by environment, and it’s difficult to solve the similarity of human motion postures and the characteristics difference of the human motion executor. This paper proposes a human posture recognition method based on few key frames sequence. Firstly, the original motion sequence is pre-selected. The initial key frame sequence is constructed by taking the extremum of the motion trajectories, and the final key frames sequence is obtained by using frame subtraction algorithm. Then, we built the hidden Markov model for different human postures and trained the model. The Baum-Welch algorithm is used to calculate the initial probability matrix, the confusion matrix and the state transition matrix, and the post-training model is obtained. Finally, the probabilities for each model are achieved by inputting the measured data and applying forward algorithm, and the gestures corresponding to the maximum probability are compared and selected as what is identified. Experiment results show that our method can efficiently select key frames of the original motion sequence, and effectively improve the accuracy of human body gesture recognition.

Key words: human posture recognition, frames sequence, frame subtraction, hidden Markov model