Research And Implementation Of Behavior And Gesture Recognition Based On Mobile Phone Sensors

The research of behavior recognition and gesture recognition based on mobile phone sensors has important theoretical and practical values.Behavior recognition has been widely used in the elderly monitoring,nursing,and user health status assessment.Gesture recognition is widely used in communication between deaf-mute people and surrounding healthy people,smart home,and virtual technology.However,there are some deficiencies in the field of existing behavior and gesture recognition.For example,behavior recognition is mostly based on wearable devices or requires sensor devices to be bound in fixed positions.In gesture recognition,there are problems such as poor anti-interference performance of the gesture interception algorithm,non-generalization of the template,precision effect due to change of the grip angle,and poor practicality of the single gesture recognition.Aiming at the above problems,this paper proposes an algorithm based on mobile phone sensors to recognize pedestrian movement state.The algorithm can identify five kinds of walk states: static,slow,medium,fast,and running,as well as stairs,escalators,elevators,upstairs,and stairs,escalators,and elevators.Pedestrian motion state recognition first carries out feature extraction of triaxial acceleration signals through wavelet transform;then uses singular value decomposition to reduce dimension of feature values;finally uses decision tree algorithm for classification recognition.The identification of the way of going upstairs and downstairs is identified by a second-level classification.The first one uses a linear fitting to perform a function fitting on the altitude data obtained by barometric pressure to distinguish the staircase or escalator upstairs,the elevator upstairs,the stairs or escalators downstairs,and the elevator downstairs;the second uses a triaxial acceleration variance threshold to distinguish between stairs and escalators,identifying six types of landings.In gesture recognition,an improved gesture recognition algorithm is proposed,which can recognize single and continuous digital and alphabetic gestures.Recognition of gestures is based on the built-in three-axis accelerometer and gyroscope.It first intercepts the gesture data through successive variance thresholds;then it resamples,filters,spline interpolations,recognizes the peaks and valleys,and normalizes the gesture data.The unified features are preprocessed;then the template training is performed based on the Dynamic Time Warping DTW matching path algorithm,the DTW algorithm based on trend matching is used for recognition,and the similar gestures in the letter recognition process are finally adopted.Acceleration displacement variation and trough sequential method are used for identification.The experimental results show that in the aspect of behavior recognition,five kinds of walking states can be identified: static,slow,medium,fast,and running,and stairs,escalators,elevators,upstairs and stairs,escalators,and elevators go downstairs and go downstairs.Gesture recognition: It is able to perform good gestures interception(eliminating hand gesture misinterpretation caused by shaking before gestures start)for user convenience;proposed cubic spline interpolation algorithm,global optimal DTW matching path model,and DTW based on trend matching The algorithm can be better identified;using acceleration integrals to differentiate the letters a,d,and q and valley order method to distinguish between i and j,eventually achieving a single gesture and the recognition of letters and continuous numbers and letters.