辅助室内定位的关键人体姿态识别

基于步行者航位推算的室内定位方法中位移的计算一定程度上依赖于人体姿态的正确识别。原地踏步和走路是其中主要的关键姿态,两者的加速度信号相似,传统方法很难进行高精度的区分,导致航位推算的步长计算错误。基于惯性传感器进行室内场景中八种人体姿态识别研究,根据运动强度的不同采用分层分类法。首先将原地踏步和走路归为一类,通过时域特征结合支持向量机(SVM)进行姿态分类;然后利用加速度的时域和小波特征以及磁场特征,结合Ada Boost方法进行二分类。关键姿态的识别准确率超过96%,对包含复杂运动姿态的步行者室内定位起到更佳的辅助作用。

Key Human Gesture Recognition for Assisting Indoor Positioning

The displacement calculation of indoor positioning method which based on pedestrian dead reckoning relies on correct human gesture identification. Marking time and walking gestures are the key gestures, and they have similar acceleration signals, which make it hard to classify them precisely by traditional methods, thus causing errors in step size calculation on dead reckoning. In this paper, eight kinds of human gestures in indoor scene are researched by using inertial sensors. Hierarchical classification means is adopted according to different motion intensity. Firstly, marking time and walking gestures are considered to be the same gesture, and then the gestures are classified by using time domain features combined with SVM. Secondly, these two gestures are classified by using acceleration time domain and wavelet features and magnetic field features combined with AdaBoost. The classification accuracy of key gestures overweighs 96%. In sum, this strategy assists pedestrian indoor positioning that incorporates complex motion well.