Human Behavior Recognition Based On Acceleration Sensor Placed Under Foot

Acceleration sensor-based human behavior recognition is a branch of behavior identification technology.It is also an important part of the artificial intelligence.Compared with the traditional,image-based behavior recognition technology,it has higher portability,lower data dimension and more robustness to interferes.This article based on the acceleration sensor which is placed under foot,our study concentrated on acceleration data acquisition system design,acceleration signal processing,the paper's main work is as follows:(1)Considering the system's portability,robustness,durability,data transmission real-time and other factors.A platform for signal acquisition,processing,analysis and display based on the Accelerometer was designed,which is placed in the insole under foot.It solves the problem of user's physical discomfort and data acquisition accuracy caused by traditional acceleration data collection mode.(2)We studied the variation of the acceleration of the feet of the human body while walking.After extracting the time domain features in the acceleration data,we proposed a dynamic balance threshold feature,it is of great importance to study the periodic movement of the foot.Then,a high-accuracy real-time step counting algorithm is designed based on the dynamic balance threshold feature.Compared with the simple threshold method and the peak detection method,this algorithm is more adaptable to multi-mode motion,and the integrated accuracy of the algorithm is 96.8%,which is better than the accuracy with mobile phone and bracelet.(3)Using support vector machine,K nearest neighbor algorithm and decision tree algorithm to classify the extracted features.After adding the dynamic balance threshold feature,the recognition accuracy of the Upslope and downhill behavior is increased by 19.93%and 20.99%respectively.Then use the principal component analysis method to reduce the dimensions of existing features,it turns out that support vector machine and decision tree algorithm are more adaptable to the characteristics after dimension reduction.