A Method of Step Counting and Behavior Recognition Based on Acceleration Sensor

Abstract

The invention discloses a step counting and motion behavior recognition method based on an acceleration sensor, which is characterized in that the sensor used in the method is placed in the following manner: the forward direction of the foot is the positive direction of the X axis, and the leftward direction is the Y axis The positive direction of the axis, the direction of lifting the foot is the negative direction of the Z axis. At the same time, due to the relatively large displacement in the X-axis direction when the human body is moving, the acceleration changes are also obvious and have obvious periodicity; therefore, as long as a reasonable threshold is set It can identify the number of moving steps and identify the movement behavior; the invention makes the waveform smoother and reduces errors by means of smoothing filter and Kalman filter, so that the system can record steps accurately in real time; at the same time, it can also accurately distinguish people's various Second, it can calculate the number of steps of various sports in real time; Third, the calculation performance of the system is not very high, and a mobile phone or a general-purpose single-chip microcomputer (MCU) with ordinary configuration can complete the calculation.

Description

Technical field [0001] The invention belongs to an acceleration sensor-based step recording and behavior recognition method. Background technique [0002] Early motion recognition was mainly based on visual methods. Given an image sequence or a video clip, the type of movement of a person was recognized. Vision-based methods have the advantages of natural interaction and rich feature information. However, this method also has some limitations in practical applications and needs to overcome many problems. Such as the lighting conditions in the environment, the position of the characters in front of the camera, the size of the venue, etc. Sensors have the advantages of being cheap, easy to carry, and not subject to site restrictions. With the development of these devices, motion recognition has been brought into a new research field, supplementing the shortcomings of traditional vision-based motion recognition methods in practical applications. , Has promoted the application of m...

AI Technical Summary

Problems solved by technology

However, in practical applications, the manual method has a poor sense of interaction and is not easy to operate and apply. Therefore, we need to perform online segmentation processing on the signal; for the selection of the classification model, most of the current research and corresponding systems use the dynamic time warping algorithm. (DTW) and hidden Markov model method (HMM), the DTW algorithm requires less training data, and can dynamically update the matching template

However, the calculation speed of this algorithm will be significantly slowed down with the length of the time series data to be recognized and the increase of the number of templates. The HMM method uses a state to represent the current action, but many whole-body actions are more complicated, and it is impossible to use only one The state is fully expressed, so two or more state variables are needed to represent it. The present invention uses the Fused HMM method to solve the problem that a single HMM cannot simultaneously model two time series sequences with correlation. The whole-body action of the process has a good description ability, and when the information of one HMM is lost, the other HMM can still work normally, which increases the robustness of the algorithm; for the problem of motion recognition in advance, the current main processing method is when a Call the recognition process after the motion is completed. In some applications, this delay will reduce the user experience.

Images