基于RSL10可穿戴设备的人体行为识别系统

Justice_Gao

基于RSL10可穿戴设备的人体行为识别系统 [复制链接]

基于RSL10可穿戴设备的人体行为识别系统

作者：Justice_Gao

一、作品简介

可穿戴设备数量在日常生活中呈指数增长，广泛应用于各个领域，例如医疗保健、生活娱乐、制造业等。利用可穿戴设备进行人体活动识别，可在不侵犯用户隐私的前提下有效地监控个人身体状况，具有很好的应用前景。本发明首先利用可穿戴设备采集人体活动是的多种传感器数据（包括加速度计、角速度计），通过蓝牙与手机建立通信连接，在手机端部署识别模型，进行人体活动识别，例如走路、跑步、坐下去、站起来、弯腰、起床、跌倒等常见活动。该项目的难点在于需要开发低功耗的数据采集系统，并获取一定量的数据，提取各类活动特征，通过机器学习方法实现有效识别。未来推广后，该项目可应用于老年人或病患安全监控、儿童异常行为识别、健身爱好者运动量评估等。

二、系统框图

如图1所示，将RSL10的纽扣传感器穿戴在身上，主要是在胸口，也可以在腰部增加一个，通过采集数据、数据处理、特征提取、训练机器学习模型、服务器部署算法等过程，通过小程序与云服务器通信，定时将用户运动时产生的六轴数据发送至服务器，并投放至存储在服务器上的机器学习算法模型中进行动作识别，最后服务器返回识别结果至小程序端，并以页面展示和语音播报两种方式呈现，以达到实时检测的效果。在用户完成运动时，将运动期间存储的六轴数据发至服务器，并投放至波峰检测代码中，根据波峰和动作的对应关系计算出每种动作的完成个数，并返回至小程序端进行页面显示。

图1 系统框图

三、部分功能说明

如图2所示，RSL10是一款超低功耗的芯片，采用ARM Cortex-M3处理器和LPDSP32 DSP内核。内部包含DC/DC电源管理模块，AES128加密引擎、DMA、88kB数据存储RAM、384kB Flash，ADC、SPI主从总线、I2C、PWM、UART、Timers，可以很好应用于可穿戴式应用。

图2 RSL10内部资源

如图3所示，BHI160集成了MCU及6轴惯性测量单元（IMU），包括一个3轴陀螺仪和3轴加速度计。Fuser Core是超低功耗32位微控制器，专为Bosch Sensortec的传感器融合和运动识别算法优化定制。与标准微控制器相比，其耗电显著降低，与Cortex M0相比省电最高达95％，与以Cortex M4为基础的设备相比省电最高达90％。

图3 BHI160传感器外设接口示意图

嵌入式软件部分的主要作用是采集加速度和角速度原始传感器，通过蓝牙上传到手机端进行算法处理，识别人体行为。系统及BHI160传感器都初始化以后，与手机建立蓝牙连接，通过蓝牙将数据传输给手机应用端。数据的第一个字节为包序列，后19个字节分别为方向、加速度、角度等数据。

图4 嵌入式软件流程图

//-----------------------------------------------------------------------------
// Copyright (c) 2018 Semiconductor Components Industries LLC
// (d/b/a "ON Semiconductor").  All rights reserved.
// This software and/or documentation is licensed by ON Semiconductor under
// limited terms and conditions.  The terms and conditions pertaining to the
// software and/or documentation are available at
// http://www.onsemi.com/site/pdf/ONSEMI_T&C.pdf ("ON Semiconductor Standard
// Terms and Conditions of Sale, Section 8 Software") and if applicable the
// software license agreement.  Do not use this software and/or documentation
// unless you have carefully read and you agree to the limited terms and
// conditions.  By using this software and/or documentation, you agree to the
// limited terms and conditions.
//-----------------------------------------------------------------------------
#include <stdio.h>
#include <math.h>

#include <BDK.h>
#include <BSP_Components.h>
#include <BLE_Components.h>
#include <SoftwareTimer.h>

#include <bsec_integration.h>

uint8_t ble_data_tx[20];
/** Converts measured orientation data into degrees.
 *
 * GREEN LED will be turned on if board heading is within 10 degrees of
 * magnetic north (<=10 or >=350).
 */
void ao_vector_cb(bhy_data_generic_t *data, bhy_virtual_sensor_t sensor)
{
    float h = data->data_vector.x / 32768.0f * 360.0f;
    float p = data->data_vector.y / 32768.0f * 360.0f;
    float y = data->data_vector.z / 32768.0f * 360.0f;

    if (h > 350.0f || h < 10.0f)
    {
        LED_On(LED_GREEN);
    }
    else
    {
        LED_Off(LED_GREEN);
    }

    ble_data_tx[1] = data->data_vector.x>>8;
    ble_data_tx[2] = data->data_vector.x;
    ble_data_tx[3] = data->data_vector.y>>8;
    ble_data_tx[4] = data->data_vector.y;
    ble_data_tx[5] = data->data_vector.z>>8;
    ble_data_tx[6] = data->data_vector.z;
    ble_data_tx[0]++;
    BLE_ICS_Notify(ble_data_tx, 20);
    printf("Orientation: h=%.1f� p=%.1f�, y=%.1f� (%d)\r\n", h, p, y, data->data_vector.status);
}

/** Converts measured linear acceleration data based on sensors dynamic range
 * and prints it to debug terminal.
 *
 * Also turns on RED LED if total acceleration applied to the board is over 1g.
 */
void la_vector_cb(bhy_data_generic_t *data, bhy_virtual_sensor_t sensor)
{
    /* Linear Acceleration sensor values are scaled using dynamic range. */
    uint16_t dyn_range = BHI160_NDOF_GetAccelDynamicRange();

    float x = data->data_vector.x / 32768.0f * dyn_range;
    float y = data->data_vector.y / 32768.0f * dyn_range;
    float z = data->data_vector.z / 32768.0f * dyn_range;

    if (fabsf(x) + fabsf(y) + fabsf(z) >= 1.0f)
    {
        LED_On(LED_RED);
    }
    else
    {
        LED_Off(LED_RED);
    }
    ble_data_tx[7] = data->data_vector.x>>8;
    ble_data_tx[8] = data->data_vector.x;
    ble_data_tx[9] = data->data_vector.y>>8;
    ble_data_tx[10] = data->data_vector.y;
    ble_data_tx[11] = data->data_vector.z>>8;
    ble_data_tx[12] = data->data_vector.z;
    ble_data_tx[0]++;
    BLE_ICS_Notify(ble_data_tx, 20);
    printf("Linear Accel: x=%.2f g, y=%.2f g, z=%.2f g\r\n", x, y, z);
}

void gyro_vector_cb(bhy_data_generic_t *data, bhy_virtual_sensor_t sensor)
{
    uint16_t dyn_range = BHI160_NDOF_GetGyroDynamicRange();

    float x = data->data_vector.x / 32768.0f * dyn_range;
    float y = data->data_vector.y / 32768.0f * dyn_range;
    float z = data->data_vector.z / 32768.0f * dyn_range;

    if (fabsf(z) > 45.0f)
    {
        LED_On(LED_BLUE);
    }
    else
    {
        LED_Off(LED_BLUE);
    }
    ble_data_tx[13] = data->data_vector.x>>8;
    ble_data_tx[14] = data->data_vector.x;
    ble_data_tx[15] = data->data_vector.y>>8;
    ble_data_tx[16] = data->data_vector.y;
    ble_data_tx[17] = data->data_vector.z>>8;
    ble_data_tx[18] = data->data_vector.z;
    ble_data_tx[0]++;
    BLE_ICS_Notify(ble_data_tx, 20);
    printf("Rate of Rotation: x=%.2f �/s, y=%.2f �/s, z=%.2f �/s\r\n", x, y, z);
}
void CustomService_WriteInd(struct BLE_ICS_RxIndData *ind)
{
    /* Print at most ind->data_len characters from ind->data array. */
    printf("APP: Received message \'%.*s\'\r\n", ind->data_len, ind->data);

    /* Echo received message back to TX characteristic. */
    //BLE_ICS_Notify(ind->data, ind->data_len);
}



int main(void)
{
    int32_t retval;
    /* Initialize all needed BDK components. */
    BDK_Initialize();
    BDK_BLE_Initialize();
    BLE_BASS_Initialize(50,1);
    BLE_ICS_Initialize(&CustomService_WriteInd);
    /* Initialize all LEDs to eliminate any currents from disabled DIO pads. */
    LED_Initialize(LED_RED);
    LED_Initialize(LED_GREEN);
    LED_Initialize(LED_BLUE);

    /* Increase I2C bus speed to 400kHz. */
    HAL_I2C_SetBusSpeed(HAL_I2C_BUS_SPEED_FAST);

    /* Initialize BHI160.
     * This may take a while depending on bus speed because RAM patch has to be
     * uploaded into BHI160 to enable support for BMM150 magnetometer.
     *
     * Set temporary timer to see how long it took to initialize and load the
     * RAM patch.
     */
    LED_On(LED_GREEN);
    {
        printf("Initializing BHI160.\r\n");

        struct SwTimer tim;
        struct SwTimer_Duration tim_dur;

        SwTimer_Initialize(&tim);
        SwTimer_Start(&tim);

        retval = BHI160_NDOF_Initialize();
        ASSERT_DEBUG(retval == BHY_SUCCESS);

        SwTimer_GetElapsed(&tim, &tim_dur);
        SwTimer_Remove(&tim);

        printf("BHI160 initialized in %lu seconds and %lu ns.\r\n",
                tim_dur.seconds, tim_dur.nanoseconds);
    }
    LED_Off(LED_GREEN);

    /* Enable desired virtual sensor outputs. */

    retval = BHI160_NDOF_EnableSensor(BHI160_NDOF_S_ORIENTATION, &ao_vector_cb, 5);
    ASSERT_DEBUG(retval == BHY_SUCCESS);

    retval = BHI160_NDOF_EnableSensor(BHI160_NDOF_S_LINEAR_ACCELERATION, &la_vector_cb, 50);
    ASSERT_DEBUG(retval == BHY_SUCCESS);

    retval = BHI160_NDOF_EnableSensor(BHI160_NDOF_S_RATE_OF_ROTATION, &gyro_vector_cb, 50);
    ASSERT_DEBUG(retval == BHY_SUCCESS);

    printf("APP: Entering main loop.\r\n");
    while (1)
    {
        BDK_Schedule();
        SYS_WAIT_FOR_INTERRUPT;
    }

    return 0;
}

数据处理算法主要对六轴原始数据进行特征提取，主要包括：

滤波：对原始数据用三阶低通滤波器滤波
FFT：对原始数据快速傅里叶变换
分割数据：对原始数据进行区间分割（将每组20s长的数据分割成多个2.56s长的小区间数据。为减小误差，分割区间时要求前后两区间具有50%重叠区，因此20s数据可以最多被分成 13 个2.56s的小区间）
特征提取：对上述每个小区间进行特征提取
保存特征：将提取特征后的特征数据进行存储，以便投入机器学习算法模型训练。