基于MEMS传感器的哑铃动作识别与能耗计算系统的设计与实现

发布时间：2018-11-23 17:05

【摘要】：为满足用户在家中进行哑铃锻炼的需求,本文基于MEMS传感器设计并实现了一个哑铃动作识别与能耗计算系统。不同于在健身房中锻炼,健身教练可给出详细的锻炼方案,并且实时纠正错误的动作,用户在家中锻炼需要自己掌握锻炼的肌肉部位,以及锻炼的效果。本系统通过采集用户哑铃锻炼的动作,识别和分析用户的动作类型、消耗能量、锻炼部位等,从而帮助用户进行正确有效的哑铃锻炼。系统分为动作采集端与计算机端两部分,动作采集端与计算机端之间通过蓝牙通信。为采集用户的运动数据,本文首先设计并实现了动作采集端。动作采集端主要负责采集运动数据,包括三维加速度、三维角速度等数据,并进行姿态解算,计算出欧拉角,为计算机端识别哑铃动作和计算哑铃运动消耗能量等工作提供数据。动作采集端由运动采集电路和普通的重量可调式哑铃组成,将运动采集电路固定于普通哑铃上,以达到采集哑铃运动数据的目的。运动采集电路主要由微处理器、运动传感器、蓝牙射频、输入输出(I/O)、电源等组成。微处理器采用了以ARM Cortex-M3为核心的STM32F103C8T6微控制器;运动传感器采用了MPU-9250九轴运动跟踪传感器,利用该传感器的DMP模块,可进行姿态解算,提供姿态数据;蓝牙射频收发采用USR-BLE101模块,该模块支持BLE 4.1协议;I/O部分包括按键、LED,以及扩展接口等;供电采用CR2032纽扣电池,通过处理器和传感器灵活的休眠与唤醒机制,可有效节约能量。动作采集端将采集到的运动数据按照预先定义的帧格式打包,通过BLE 4.1发送到计算机端。计算机端主要负责哑铃动作识别与能耗计算,包括动作识别方法和能耗计算方法两部分。本文提出了一种基于MEMS传感器的动作识别方法,该方法通过读取三维加速度与三维角速度的实时采样数据,并对数据进行加速度分解、滤波、周期判定、归一化等处理后,与特征数据库中的数据进行相似度分析,再通过比较待测动作与数据库中每组动作的皮尔逊相关系数,从而实现对特定动作的识别,并可对重复性动作进行计数。本文对6种常见哑铃动作进行了动作识别实验,验证了动作识别方法的有效性,且平均识别率达到94%。能耗计算方法对每种哑铃动作所锻炼的肌肉部位进行分析,同时通过对哑铃动能的计算,分析了每种动作的能量(热量)消耗情况,进而统计了每个肌肉部位所消耗的能量与用户消耗的总能量。通过哑铃动作识别与能耗计算系统,用户能够精确地掌握自身肌肉的锻炼情况,从而更加科学地进行哑铃锻炼。
[Abstract]:In order to meet the demand of dumbbell exercise at home, a dumbbell action recognition and energy consumption calculation system based on MEMS sensor is designed and implemented in this paper. Instead of exercising in the gym, the trainer can give a detailed exercise plan and correct the wrong action in real time. The user needs to master the muscle part of the exercise and the effect of the exercise in the home. The system collects the dumbbell movement of the user, identifies and analyzes the user's action type, energy consumption, exercise position and so on, so as to help the user to carry on the correct and effective dumbbell exercise. The system is divided into two parts: the action acquisition end and the computer terminal, and the communication between the action acquisition end and the computer end through Bluetooth. In order to collect the user's motion data, this paper first designs and implements the action acquisition terminal. The action acquisition end is mainly responsible for collecting motion data, including three-dimensional acceleration, three-dimensional angular velocity and so on, and carries out attitude calculation to calculate the Euler angle. It provides data for the computer to identify dumbbell movement and calculate dumbbell motion energy consumption. The motion acquisition terminal is composed of a motion acquisition circuit and a general weight adjustable dumbbell. The motion acquisition circuit is fixed on the ordinary dumbbell to achieve the purpose of collecting dumbbell motion data. Motion acquisition circuit consists of microprocessor, motion sensor, Bluetooth RF, I / O (I / O), power supply and so on. The microprocessor adopts the STM32F103C8T6 microcontroller with ARM Cortex-M3 as the core, the motion sensor adopts the MPU-9250 nine-axis motion tracking sensor, and the DMP module of the sensor can be used to calculate the attitude and provide the attitude data. Bluetooth RF transceiver adopts USR-BLE101 module, which supports BLE 4.1 protocol, I / O part includes keystroke, LED, and extended interface, etc. CR2032 button battery is used for power supply, which can save energy effectively through flexible dormancy and wake-up mechanism of processor and sensor. The motion data is packed in the predefined frame format and sent to the computer through BLE 4.1. The computer is mainly responsible for dumbbell action recognition and energy consumption calculation, including motion recognition method and energy consumption calculation method. In this paper, an action recognition method based on MEMS sensor is proposed. After reading the real time sampling data of 3D acceleration and 3D angular velocity, the data are processed by acceleration decomposition, filtering, period determination, normalization and so on. By analyzing the similarity with the data in the feature database and comparing the Pearson correlation coefficient between the actions to be tested and each group of actions in the database, the recognition of specific actions can be realized, and the repetitive actions can be counted. In this paper, six kinds of common dumbbell actions are tested, and the results show that the method is effective, and the average recognition rate is 94. The energy consumption calculation method is used to analyze the muscle parts of each dumbbell movement. At the same time, the energy (heat) consumption of each action is analyzed by calculating the kinetic energy of the dumbbell. Furthermore, the energy consumed by each muscle part and the total energy consumed by the user are counted. Through the dumbbell action recognition and energy consumption calculation system, the user can accurately grasp the exercise of their muscles, thus more scientific dumbbell exercise.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9

【相似文献】