monster-little

AI健身计数

AI健身计数

一、jupyter notebook部分

（姿势估计 mediapipe）

1、环境的配置

下载mediapipe包

pip install mediapipe==0.8.9 -i https://mirrors.aliyun.com/pypi/simple/

下载open-cv包

pip install opencv-python -i https://mirrors.aliyun.com/pypi/simple/

2、单张图片的识别

# opencv-python
import cv2
# mediapipe人工智能工具包
import mediapipe as mp
# 进度条库
from tqdm import tqdm
# 时间库
import time
# 导入python绘图matplotlib
import matplotlib.pyplot as plt
# 使用ipython的魔法方法，将绘制出的图像直接嵌入在notebook单元格中
%matplotlib inline

# 定义可视化图像函数
def look_img(img):
    '''opencv读入图像格式为BGR，matplotlib可视化格式为RGB，因此需将BGR转RGB'''
    img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    plt.imshow(img_RGB)
    plt.show()
    
# 导入solution
mp_pose = mp.solutions.pose

# # 导入绘图函数
mp_drawing = mp.solutions.drawing_utils 

# 导入模型
pose = mp_pose.Pose(static_image_mode=True,        # 是静态图片还是连续视频帧
                    model_complexity=1,            # 选择人体姿态关键点检测模型，0最快但性能差，2性能好但是慢
                    smooth_landmarks=True,         # 是否平滑关键点
                    min_detection_confidence=0.5,  # 置信度阈值
                    min_tracking_confidence=0.5)   # 追踪阈值

# 从图片文件读入图像，opencv读入为BGR格式
img = cv2.imread('person1.jfif')
# BGR转RGB
img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# 将RGB图像输入模型，获取预测结果
results = pose.process(img_RGB)

# 可视化
mp_drawing.draw_landmarks(img, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
look_img(img)
# 在三维真实物理坐标系中可视化以米为单位的检测结果
mp_drawing.plot_landmarks(results.pose_world_landmarks, mp_pose.POSE_CONNECTIONS)

3、 摄像头实时检测

# opencv-python
import cv2
# mediapipe人工智能工具包
import mediapipe as mp
# 进度条库
from tqdm import tqdm
# 时间库
import time
# 导入solution
mp_pose = mp.solutions.pose

# # 导入绘图函数
mp_drawing = mp.solutions.drawing_utils 

# 导入模型
pose = mp_pose.Pose(static_image_mode=False,        # 是静态图片还是连续视频帧
                    model_complexity=2,            # 选择人体姿态关键点检测模型，0性能差但快，2性能好但慢，1介于两者之间
                    smooth_landmarks=True,         # 是否平滑关键点
                    enable_segmentation=True,      # 是否人体抠图
                    min_detection_confidence=0.5,  # 置信度阈值
                    min_tracking_confidence=0.5)   # 追踪阈值

# 处理帧函数
def process_frame(img):
    # BGR转RGB
    img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    # 将RGB图像输入模型，获取预测结果
    results = pose.process(img_RGB)

    # 可视化
    mp_drawing.draw_landmarks(img, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
    # look_img(img)
    # 在三维真实物理坐标系中可视化以米为单位的检测结果
    # mp_drawing.plot_landmarks(results.pose_world_landmarks, mp_pose.POSE_CONNECTIONS)
    
#     # BGR转RGB
#     img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
#     # 将RGB图像输入模型，获取预测结果
#     results = hands.process(img_RGB)
    
#     if results.multi_hand_landmarks: # 如果有检测到手
#         # 遍历每一只检测出的手
#         for hand_idx in range(len(results.multi_hand_landmarks)):
#             hand_21 = results.multi_hand_landmarks[hand_idx] # 获取该手的所有关键点坐标
#             mpDraw.draw_landmarks(img, hand_21, mp_hands.HAND_CONNECTIONS) # 可视化
            
    return img


# 调用摄像头逐帧实时处理模板
# 不需修改任何代码，只需修改process_frame函数即可
# 同济子豪兄 2021-7-8

# 导入opencv-python
import cv2
import time

# 获取摄像头，传入0表示获取系统默认摄像头
cap = cv2.VideoCapture(1)

# 打开cap
cap.open(0)

# 无限循环，直到break被触发
while cap.isOpened():
    # 获取画面
    success, frame = cap.read()
    if not success:
        print('Error')
        break
    
    ## !!!处理帧函数
    frame = process_frame(frame)
    
    # 展示处理后的三通道图像
    cv2.imshow('my_window',frame)

    if cv2.waitKey(1) in [ord('q'),27]: # 按键盘上的q或esc退出（在英文输入法下）
        break
    
# 关闭摄像头
cap.release()

# 关闭图像窗口
cv2.destroyAllWindows()

英文状态下按q退出

可以运行 就不放丑照了

4、摄像头实时检测的提升

（不同身体部位加入了不同颜色的结点）

# opencv-python
import cv2
# mediapipe人工智能工具包
import mediapipe as mp
# 进度条库
from tqdm import tqdm
# 时间库
import time
# 导入solution
mp_pose = mp.solutions.pose

# # 导入绘图函数
mp_drawing = mp.solutions.drawing_utils 

# 导入模型
pose = mp_pose.Pose(static_image_mode=False,        # 是静态图片还是连续视频帧
                    model_complexity=2,            # 选择人体姿态关键点检测模型，0性能差但快，2性能好但慢，1介于两者之间
                    smooth_landmarks=True,         # 是否平滑关键点
                    min_detection_confidence=0.5,  # 置信度阈值
                    min_tracking_confidence=0.5)   # 追踪阈值
                    
def process_frame(img):
    
    # 记录该帧开始处理的时间
    start_time = time.time()
    
    # 获取图像宽高
    h, w = img.shape[0], img.shape[1]

    # BGR转RGB
    img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    # 将RGB图像输入模型，获取预测结果
    results = pose.process(img_RGB)

    if results.pose_landmarks: # 若检测出人体关键点

        # 可视化关键点及骨架连线
        mp_drawing.draw_landmarks(img, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

        for i in range(33): # 遍历所有33个关键点，可视化

            # 获取该关键点的三维坐标
            cx = int(results.pose_landmarks.landmark[i].x * w)
            cy = int(results.pose_landmarks.landmark[i].y * h)
            cz = results.pose_landmarks.landmark[i].z

            radius = 10

            if i == 0: # 鼻尖
                img = cv2.circle(img,(cx,cy), radius, (0,0,255), -1)
            elif i in [11,12]: # 肩膀
                img = cv2.circle(img,(cx,cy), radius, (223,155,6), -1)
            elif i in [23,24]: # 髋关节
                img = cv2.circle(img,(cx,cy), radius, (1,240,255), -1)
            elif i in [13,14]: # 胳膊肘
                img = cv2.circle(img,(cx,cy), radius, (140,47,240), -1)
            elif i in [25,26]: # 膝盖
                img = cv2.circle(img,(cx,cy), radius, (0,0,255), -1)
            elif i in [15,16,27,28]: # 手腕和脚腕
                img = cv2.circle(img,(cx,cy), radius, (223,155,60), -1)
            elif i in [17,19,21]: # 左手
                img = cv2.circle(img,(cx,cy), radius, (94,218,121), -1)
            elif i in [18,20,22]: # 右手
                img = cv2.circle(img,(cx,cy), radius, (16,144,247), -1)
            elif i in [27,29,31]: # 左脚
                img = cv2.circle(img,(cx,cy), radius, (29,123,243), -1)
            elif i in [28,30,32]: # 右脚
                img = cv2.circle(img,(cx,cy), radius, (193,182,255), -1)
            elif i in [9,10]: # 嘴
                img = cv2.circle(img,(cx,cy), radius, (205,235,255), -1)
            elif i in [1,2,3,4,5,6,7,8]: # 眼及脸颊
                img = cv2.circle(img,(cx,cy), radius, (94,218,121), -1)
            else: # 其它关键点
                img = cv2.circle(img,(cx,cy), radius, (0,255,0), -1)

        # 展示图片
        # look_img(img)

    else:
        scaler = 1
        failure_str = 'No Person'
        img = cv2.putText(img, failure_str, (25 * scaler, 100 * scaler), cv2.FONT_HERSHEY_SIMPLEX, 1.25 * scaler, (255, 0, 255), 2 * scaler)
        # print('从图像中未检测出人体关键点，报错。')
        
    # 记录该帧处理完毕的时间
    end_time = time.time()
    # 计算每秒处理图像帧数FPS
    FPS = 1/(end_time - start_time)
    
    scaler = 1
    # 在图像上写FPS数值，参数依次为：图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细
    img = cv2.putText(img, 'FPS  '+str(int(FPS)), (25 * scaler, 50 * scaler), cv2.FONT_HERSHEY_SIMPLEX, 1.25 * scaler, (255, 0, 255), 2 * scaler)
    return img

# 调用摄像头逐帧实时处理模板
# 不需修改任何代码，只需定义process_frame函数即可
# 同济子豪兄 2021-7-8

# 导入opencv-python
import cv2
import time

# 获取摄像头，传入0表示获取系统默认摄像头
cap = cv2.VideoCapture(1)

# 打开cap
cap.open(0)

# 无限循环，直到break被触发
while cap.isOpened():
    # 获取画面
    success, frame = cap.read()
    if not success:
        break
    
    ## !!!处理帧函数
    frame = process_frame(frame)
    
    # 展示处理后的三通道图像
    cv2.imshow('my_window', frame)

    if cv2.waitKey(1) in [ord('q'),27]: # 按键盘上的q或esc退出（在英文输入法下）
        break
    
# 关闭摄像头
cap.release()

# 关闭图像窗口
cv2.destroyAllWindows()

可以运行 不放丑照了

5、视频处理

# opencv-python
import cv2
# mediapipe人工智能工具包
import mediapipe as mp
# 进度条库
from tqdm import tqdm
# 时间库
import time
# 导入solution
mp_pose = mp.solutions.pose

# # 导入绘图函数
mp_drawing = mp.solutions.drawing_utils 

# 导入模型
pose = mp_pose.Pose(static_image_mode=False,        # 是静态图片还是连续视频帧
                    model_complexity=2,            # 选择人体姿态关键点检测模型，0性能差但快，2性能好但慢，1介于两者之间
                    smooth_landmarks=True,         # 是否平滑关键点
                    min_detection_confidence=0.5,  # 置信度阈值
                    min_tracking_confidence=0.5)   # 追踪阈值
def process_frame(img):
    
    '''
    B站：同济子豪兄（https://space.bilibili.com/1900783）

    微信公众号：人工智能小技巧
    '''
    
    # 记录该帧开始处理的时间
    start_time = time.time()
    
    # 获取图像宽高
    h, w = img.shape[0], img.shape[1]

    # BGR转RGB
    img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    # 将RGB图像输入模型，获取预测结果
    results = pose.process(img_RGB)

    if results.pose_landmarks: # 若检测出人体关键点

        # 可视化关键点及骨架连线
        mp_drawing.draw_landmarks(img, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

        for i in range(33): # 遍历所有33个关键点，可视化

            # 获取该关键点的三维坐标
            cx = int(results.pose_landmarks.landmark[i].x * w)
            cy = int(results.pose_landmarks.landmark[i].y * h)
            cz = results.pose_landmarks.landmark[i].z

            radius = 10

            if i == 0: # 鼻尖
                img = cv2.circle(img,(cx,cy), radius, (0,0,255), -1)
            elif i in [11,12]: # 肩膀
                img = cv2.circle(img,(cx,cy), radius, (223,155,6), -1)
            elif i in [23,24]: # 髋关节
                img = cv2.circle(img,(cx,cy), radius, (1,240,255), -1)
            elif i in [13,14]: # 胳膊肘
                img = cv2.circle(img,(cx,cy), radius, (140,47,240), -1)
            elif i in [25,26]: # 膝盖
                img = cv2.circle(img,(cx,cy), radius, (0,0,255), -1)
            elif i in [15,16,27,28]: # 手腕和脚腕
                img = cv2.circle(img,(cx,cy), radius, (223,155,60), -1)
            elif i in [17,19,21]: # 左手
                img = cv2.circle(img,(cx,cy), radius, (94,218,121), -1)
            elif i in [18,20,22]: # 右手
                img = cv2.circle(img,(cx,cy), radius, (16,144,247), -1)
            elif i in [27,29,31]: # 左脚
                img = cv2.circle(img,(cx,cy), radius, (29,123,243), -1)
            elif i in [28,30,32]: # 右脚
                img = cv2.circle(img,(cx,cy), radius, (193,182,255), -1)
            elif i in [9,10]: # 嘴
                img = cv2.circle(img,(cx,cy), radius, (205,235,255), -1)
            elif i in [1,2,3,4,5,6,7,8]: # 眼及脸颊
                img = cv2.circle(img,(cx,cy), radius, (94,218,121), -1)
            else: # 其它关键点
                img = cv2.circle(img,(cx,cy), radius, (0,255,0), -1)

        # 展示图片
        # look_img(img)

    else:
        scaler = 1
        failure_str = 'No Person'
        img = cv2.putText(img, failure_str, (25 * scaler, 100 * scaler), cv2.FONT_HERSHEY_SIMPLEX, 1.25 * scaler, (255, 0, 255), 2 * scaler)
        # print('从图像中未检测出人体关键点，报错。')
        
    # 记录该帧处理完毕的时间
    end_time = time.time()
    # 计算每秒处理图像帧数FPS
    FPS = 1/(end_time - start_time)
    
    scaler = 1
    # 在图像上写FPS数值，参数依次为：图片，添加的文字，左上角坐标，字体，字体大小，颜色，字体粗细
    img = cv2.putText(img, 'FPS  '+str(int(FPS)), (25 * scaler, 50 * scaler), cv2.FONT_HERSHEY_SIMPLEX, 1.25 * scaler, (255, 0, 255), 2 * scaler)
    return img

# 视频逐帧处理代码模板
# 不需修改任何代码，只需定义process_frame函数即可
# 同济子豪兄 2021-7-10

def generate_video(input_path='./videos/three-hands.mp4'):
    filehead = input_path.split('/')[-1]
    output_path = "out-" + filehead
    
    print('视频开始处理',input_path)
    
    # 获取视频总帧数
    cap = cv2.VideoCapture(input_path)
    frame_count = 0
    while(cap.isOpened()):
        success, frame = cap.read()
        frame_count += 1
        if not success:
            break
    cap.release()
    print('视频总帧数为',frame_count)
    
    # cv2.namedWindow('Crack Detection and Measurement Video Processing')
    cap = cv2.VideoCapture(input_path)
    frame_size = (cap.get(cv2.CAP_PROP_FRAME_WIDTH), cap.get(cv2.CAP_PROP_FRAME_HEIGHT))

    # fourcc = int(cap.get(cv2.CAP_PROP_FOURCC))
    # fourcc = cv2.VideoWriter_fourcc(*'XVID')
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    fps = cap.get(cv2.CAP_PROP_FPS)

    out = cv2.VideoWriter(output_path, fourcc, fps, (int(frame_size[0]), int(frame_size[1])))
    
    # 进度条绑定视频总帧数
    with tqdm(total=frame_count-1) as pbar:
        try:
            while(cap.isOpened()):
                success, frame = cap.read()
                if not success:
                    break

                # 处理帧
                # frame_path = './temp_frame.png'
                # cv2.imwrite(frame_path, frame)
                try:
                    frame = process_frame(frame)
                except:
                    print('error')
                    pass
                
                if success == True:
                    # cv2.imshow('Video Processing', frame)
                    out.write(frame)

                    # 进度条更新一帧
                    pbar.update(1)

                # if cv2.waitKey(1) & 0xFF == ord('q'):
                    # break
        except:
            print('中途中断')
            pass

    cv2.destroyAllWindows()
    out.release()
    cap.release()
    print('视频已保存', output_path)
    
generate_video(input_path='./badminton.mp4')

（此处找了一个打羽毛球的视频）

二、Google Colabor部分

（示例深蹲）（需要进入谷歌）

1、数据集

squat_dataset

  1. down

  1. up

2、数据集处理

  1. 对每个图片进行处理 对每个图片识别出有33个关节点

  1. 每个点在3维坐标系下有x y z ，生成的csv中 每个图片有3*33=99个数据

  1. 对视频进行预测

三、 安卓开发部分

1、导入项目

2、添加一个device

（API选择30 选择高级设置下面的摄像机为webcam0）

3、运行

点击运行键

注意：在右上角设置中 打开深蹲计数器

代码来自同济自豪兄 侵权删

posted on 2022-07-31 10:52  monster-little  阅读(406)  评论(0)  编辑  收藏  举报

导航