Research On Non-contact Sensing Method Of Human Micro-motion

Non-contact sensing of human motion by wireless signals such as millimeter wave(mmWave)and acoustic wave is a new frontier research hot-spot in recent years.At present,human motion sensing usually focuses on the body’s trunk,limbs,and other large-scale movements,but the human body also has local micro-movements such as the finger,chest fluctuation.The non-contact human micro-motion sensing will revolutionize the development of human-computer interaction,health monitoring,and other fields,and has important research significance and application value.This article focuses on three important challenges of non-contact sensing of human micro-motion:distinguishing sensing of similar micro-motions,universal sensing of crossindividual micro-motions,depth sensing of micro-motions representation information.The following micro-motion intelligent sensing models and methods based on acoustic and mm Wave are proposed and applied to handwriting input based on finger trajectory micro-motion and emotion recognition based on chest micro-vibration.The main contributions of this article are as follows:(1)Similar micro-motion sensing method based on motion trajectory features.Due to the small displacement of micro-motion,the trajectories of similar micro-motions are similar and difficult to distinguish.To solve the distinctively sensing of similar micro-motions,this article leverages the inherent invariance characteristics of the motion trajectory and the contextual relationship of the micro-motion in the environment to explore an effective motion recognition method.This article proposes a feature-preserved fast alignment mechanism,which can efficiently extract features that characterize trajectory patterns to calculate the similarity between micro-motion trajectories with different lengths.According to the contextual relationship of multiple continuous micro-motions,this article also designs a Bayesian probability model based on scoring to realize the precise continuous micro-motions sensing.And a micro-motion training templates pre-selection method is designed to ensure the real-time performance of the system.Experimental results show that compared with existing methods,this method can achieve more accurate micro-motion recognition in real-time.(2)Cross-individual micro-motion sensing method based on deep neural network.Due to different users having different behavior habits,micromotions have individual differences.To solve universal sensing of cross individual micro-motion,this article designs a neural network called InceptionLSTM,in which the Inception module extracts universal depth local features of the same micro-motion;the LSTM module models the time-series relationship between the frames of the micro-motion signal to build a universal model suitable for multiple users.In addition,according to the contextual relationship between micro-motions mined from a large number of continuous micro-motion samples,this article establishes a multi-class bi-gram language model to realize continuous micro-motions sensing.Experimental results show that this method can achieve high accuracy of continuous micro-motion recognition for users without training.(3)Depth sensing of micro-motion based on multi-scale attention model.For some micro-motions,their behavior patterns can reflect the person’s state and even emotions.To solve the depth sensing of micro-motion,this article first conducts a measurement study to show the feasibility of using depth information-related features to achieve depth information sensing.Further,for the problem of the selection of effective features in complex environments and the time-varying deviation correction of features,a multi-scale network structure based on the feature attention mechanism is designed,in which the multi-scale fusion module is used to obtain richer feature expression,and the attention module is used to strengthen important features and weaken invalid features.Experimental results show that the proposed method can establish a highly robust mapping between micro-motions and people/depth information,and achieve accurate user identification and emotion sensing.(4)Prototype systems of contactless handwriting input and emotion recognition.To verify the effectiveness of the proposed algorithm,two prototype systems are designed and implemented:the handwriting input system and the emotion recognition system.A large number of system experiments show that the methods proposed in this article can effectively distinguish similar micromotions and establish a universal model for different users,and mine the depth information behind micro-motion.These two systems can be used in many fields such as human-computer interaction and health monitoring.To sum up,this article explores the key challenges of non-contact sensing of human micro-motion from three aspects:distinction sensing of similar micro-motions,universal sensing of cross-individual micro-motions,and depth sensing of micro-motions.A series of model algorithms are proposed and a prototype system is constructed for evaluation,which provides the theoretical basis and technical support for the future development of non-contact sensing of human micro-motion.