ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (3): 621-631.doi: 10.7544/issn1000-1239.2016.20148159

• 人工智能 • 上一篇    下一篇



  1. (东华大学计算机科学与技术学院 上海 201620) (
  • 出版日期: 2016-03-01
  • 基金资助: 

Human Motion Recognition Based on Triaxial Accelerometer


  1. (College of Computer Science and Technology, Donghua University, Shanghai 201620)
  • Online: 2016-03-01

摘要: 提出并实现一种基于移动设备的用户运动行为的检测算法.在用户随身携带移动设备的情况下,算法就可以根据移动设备中的三轴加速度数据,判别出来用户的行为状态.算法综合分析了加速度传感器数据的时域和频域特性,并通过方向无关性和步幅处理,进一步提高算法的适应性.算法对所抽取21个运动特征值进行了主成分分析,找出了11个主要特征成分,然后使用这些主成分对运行数据进行识别分类.提高了算法准确度,并降低了算法的时间和空间复杂度.在对分类算法综合分析和比较后,J48判决树算法被采纳.算法还根据人类运动的习惯和特性,对特性分类并计算分类的结果,再采用隐式Markov模型进行处理,进一步提高识别的准确度.对多人、多状态数据的实验表明,这种综合方法具有较高的识别准确度和适应性,在对多人多次实际运动数据的处理中,正确识别率可以达到96.13%.

关键词: 人体运动识别, 运动分析, 主成分分析, 隐Markov模型, 判决树

Abstract: In this paper, a new method is presented to realize motion detection on a mobile device. The scheme can recognize the people’s motions state according to the acceleration data as long as they simply carry a mobile device with a build-in triaxial accelerometer. The features of the motion signal are extracted in frequency domain and time domain using the method of comprehensive analysis. To enhance the adaptability of the method, the algorithm of independent direction of mobile device algorithm has been applied. The 11 major components, which have greatest contribution to the motion detection, are selected from the 21 motion’s features by principal component analysis, so the input dimension is reduced and the computational complexity of time and space of the algorithm is decreased. Based on the analysis and synthetic comparison of various classification algorithm, the J48 decision tree is accepted. According to the characteristics of the people nature motion, the hidden Markov model is introduced to improve the detection accuracy. Experiments, with different person and different motion, show that the synthesis algorithm has good accuracy and adaptability, and the highest recognition rate achieves 96.13%.

Key words: human motion recognition, motion analysis, principal component analysis, hidden Markov model, decision tree