MediaPipe实现手指关键点检测及追踪,人脸识别及追踪

Python之王

共 17918字,需浏览 36分钟

 ·

2022-01-03 02:53

OpenCV 是一个用于计算机视觉应用程序的库。在 OpenCV 的帮助下,我们可以构建大量实时运行更好的应用程序。主要用于图像和视频处理。

除了 OpenCV,我们将使用 MediaPipe 库。

1.MediaPipe简介

MediaPipe是一个主要用于构建音频、视频或任何时间序列数据的框架。在 MediaPipe 框架的帮助下,我们可以为不同的媒体处理功能构建管道。

MediaPipe 的一些主要应用。

  • 多手追踪 -  人脸检测 -  对象检测和跟踪 -  Objection:3D 对象检测和跟踪 -  AutoFlip:自动视频裁剪管道等。MediaPipe 使用单次手掌检测模型,一旦完成,它会对检测到的手部区域中的 21 个 3D 手掌坐标执行精确的关键点定位。

MediaPipe 管道使用多个模型,例如,从完整图像返回定向手边界框的手掌检测模型。裁剪后的图像区域被馈送到由手掌检测器定义的手部标志模型,并返回高保真 3D 手部关键点。

现在让我们实现手部跟踪模型。

安装所需的模块

–> pip install opencv-python

–> pip install mediapipe

注意:这里的python版本尽量在3.8以上,不然会报各种错误!!

首先,让我们检查网络摄像头的工作情况。

import cv2
import mediapipe as mp
import time

cap = cv2.VideoCapture(0)

mpHands = mp.solutions.hands
hands = mpHands.Hands(static_image_mode=False,
                      max_num_hands=2,
                      min_detection_confidence=0.5,
                      min_tracking_confidence=0.5)
mpDraw = mp.solutions.drawing_utils

pTime = 0
cTime = 0

while True:
    success, img = cap.read()
    img = cv2.flip(img, 1)
    imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    results = hands.process(imgRGB)
    #print(results.multi_hand_landmarks)

    if results.multi_hand_landmarks:
        for handLms in results.multi_hand_landmarks:
            for id, lm in enumerate(handLms.landmark):
                #print(id,lm)
                h, w, c = img.shape
                cx, cy = int(lm.x *w), int(lm.y*h)
                #if id ==0:
                cv2.circle(img, (cx,cy), 3, (255,0,255), cv2.FILLED)

            mpDraw.draw_landmarks(img, handLms, mpHands.HAND_CONNECTIONS)


    cTime = time.time()
    fps = 1/(cTime-pTime)
    pTime = cTime

    cv2.putText(img,str(int(fps)), (10,70), cv2.FONT_HERSHEY_PLAIN, 3, (255,0,255), 3)

    cv2.imshow("Image", img)
    cv2.waitKey(1)

结果见图示:

2.mediapipe实现手掌拖拽屏幕中的方块(转载于恩培大佬)

效果如示:

源码如下:

"""
功能:手势虚拟拖拽
1、使用OpenCV读取摄像头视频流;
2、识别手掌关键点像素坐标;
3、根据食指和中指指尖的坐标,利用勾股定理计算距离,当距离较小且都落在矩形内,则触发拖拽(矩形变色);
4、矩形跟着手指动;
5、两指放开,则矩形停止移动
"
""


# 导入OpenCV
import cv2
# 导入mediapipe
import mediapipe as mp

# 导入其他依赖包
import time
import math



# 方块管理类
class SquareManager:
    def __init__(self, rect_width):
        
        
        # 方框长度
        self.rect_width = rect_width
        
        # 方块list
        self.square_count = 0
        self.rect_left_x_list = []
        self.rect_left_y_list = []
        self.alpha_list = []

        # 中指与矩形左上角点的距离
        self.L1 = 0
        self.L2 = 0

        # 激活移动模式
        self.drag_active = False

        # 激活的方块ID
        self.active_index = -1
        
    
    # 创建一个方块,但是没有显示
    def create(self,rect_left_x,rect_left_y,alpha=0.4):
        self.rect_left_x_list.append(rect_left_x)
        self.rect_left_y_list.append(rect_left_y)
        self.alpha_list.append(alpha)
        self.square_count+=1
        

    # 更新位置
    def display(self,class_obj):
        for i in range(0,self.square_count):
            x= self.rect_left_x_list[i]
            y= self.rect_left_y_list[i]
            alpha  = self.alpha_list[i]

            overlay = class_obj.image.copy()

            if(i == self.active_index):
                cv2.rectangle(overlay,(x,y),(x+self.rect_width,y+self.rect_width),(255, 0, 255),-1)
            else:
                cv2.rectangle(overlay,(x,y),(x+self.rect_width,y+self.rect_width),(255, 0, 0),-1)
            
            # Following line overlays transparent rectangle over the self.image
            class_obj.image = cv2.addWeighted(overlay, alpha, class_obj.image, 1 - alpha, 0)
            


    # 判断落在哪个方块上,返回方块的ID
    def checkOverlay(self,check_x,check_y):
        for i in range(0,self.square_count):
            x= self.rect_left_x_list[i]
            y= self.rect_left_y_list[i]

            if (x <  check_x < (x+self.rect_width) ) and ( y < check_y < (y+self.rect_width)):
                
                # 保存被激活的方块ID
                self.active_index = i

                return i
        
        return -1

    # 计算与指尖的距离
    def setLen(self,check_x,check_y):
        # 计算距离
        self.L1 = check_x - self.rect_left_x_list[self.active_index] 
        self.L2 = check_y - self.rect_left_y_list[self.active_index]

    # 更新方块    
    def updateSquare(self,new_x,new_y):
        # print(self.rect_left_x_list[self.active_index])
        self.rect_left_x_list[self.active_index] = new_x - self.L1 
        self.rect_left_y_list[self.active_index] = new_y - self.L2


# 识别控制类
class HandControlVolume:
    def __init__(self):
        # 初始化medialpipe
        self.mp_drawing = mp.solutions.drawing_utils
        self.mp_drawing_styles = mp.solutions.drawing_styles
        self.mp_hands = mp.solutions.hands

        
        # 中指与矩形左上角点的距离
        self.L1 = 0
        self.L2 = 0

        # image实例,以便另一个类调用
        self.image=None

        

    # 主函数
    def recognize(self):
        # 计算刷新率
        fpsTime = time.time()

        # OpenCV读取视频流
        cap = cv2.VideoCapture(0)
        # 视频分辨率
        resize_w = 1280
        resize_h = 720

        # 画面显示初始化参数
        rect_percent_text = 0


        # 初始化方块管理器
        squareManager = SquareManager(150)

        # 创建多个方块
        for i in range(0, 7):
            squareManager.create(180*i+20, 180, 0.3) ##原来是0.6


        with self.mp_hands.Hands(min_detection_confidence=0.7,
                                 min_tracking_confidence=0.5,
                                 max_num_hands=2) as hands:
            while cap.isOpened():

                # 初始化矩形
                success, self.image = cap.read()
                self.image = cv2.resize(self.image, (resize_w, resize_h))

                if not success:
                    print("空帧.")
                    continue

                # 提高性能
                self.image.flags.writeable = False
                # 转为RGB
                self.image = cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
                # 镜像
                self.image = cv2.flip(self.image, 1)
                # mediapipe模型处理
                results = hands.process(self.image)

                self.image.flags.writeable = True
                self.image = cv2.cvtColor(self.image, cv2.COLOR_RGB2BGR)
                # 判断是否有手掌
                if results.multi_hand_landmarks:
                    # 遍历每个手掌
                    for hand_landmarks in results.multi_hand_landmarks:
                        # 在画面标注手指
                        self.mp_drawing.draw_landmarks(
                            self.image,
                            hand_landmarks,
                            self.mp_hands.HAND_CONNECTIONS,
                            self.mp_drawing_styles.get_default_hand_landmarks_style(),
                            self.mp_drawing_styles.get_default_hand_connections_style())


                        # 解析手指,存入各个手指坐标
                        landmark_list = []

                        # 用来存储手掌范围的矩形坐标
                        paw_x_list = []
                        paw_y_list = []
                        for landmark_id, finger_axis in enumerate(
                                hand_landmarks.landmark):
                            landmark_list.append([
                                landmark_id, finger_axis.x, finger_axis.y,
                                finger_axis.z
                            ])
                            paw_x_list.append(finger_axis.x)
                            paw_y_list.append(finger_axis.y)
                        if landmark_list:
                            # 比例缩放到像素
                            ratio_x_to_pixel = lambda x: math.ceil(x * resize_w)
                            ratio_y_to_pixel = lambda y: math.ceil(y * resize_h)
                            
                            # 设计手掌左上角、右下角坐标
                            paw_left_top_x,paw_right_bottom_x = map(ratio_x_to_pixel,[min(paw_x_list),max(paw_x_list)])
                            paw_left_top_y,paw_right_bottom_y = map(ratio_y_to_pixel,[min(paw_y_list),max(paw_y_list)])

                            # 给手掌画框框
                            cv2.rectangle(self.image,(paw_left_top_x-30,paw_left_top_y-30),(paw_right_bottom_x+30,paw_right_bottom_y+30),(0, 255,0),2)

                            # 获取中指指尖坐标
                            middle_finger_tip = landmark_list[12]
                            middle_finger_tip_x =ratio_x_to_pixel(middle_finger_tip[1])
                            middle_finger_tip_y =ratio_y_to_pixel(middle_finger_tip[2])

                            # 获取食指指尖坐标
                            index_finger_tip = landmark_list[8]
                            index_finger_tip_x =ratio_x_to_pixel(index_finger_tip[1])
                            index_finger_tip_y =ratio_y_to_pixel(index_finger_tip[2])
                            # 中间点
                            between_finger_tip = (middle_finger_tip_x+index_finger_tip_x)//2, (middle_finger_tip_y+index_finger_tip_y)//2
                            # print(middle_finger_tip_x)
                            thumb_finger_point = (middle_finger_tip_x,middle_finger_tip_y)
                            index_finger_point = (index_finger_tip_x,index_finger_tip_y)
                            # 画指尖2点
                            circle_func = lambda point: cv2.circle(self.image,point,10,(255,0,255),-1)
                            self.image = circle_func(thumb_finger_point)
                            self.image = circle_func(index_finger_point)
                            self.image = circle_func(between_finger_tip)
                            # 画2点连线
                            self.image = cv2.line(self.image,thumb_finger_point,index_finger_point,(255,0,255),5)
                            # 勾股定理计算长度
                            line_len = math.hypot((index_finger_tip_x-middle_finger_tip_x),(index_finger_tip_y-middle_finger_tip_y))
                            # 将指尖距离映射到文字
                            rect_percent_text = math.ceil(line_len)

                            # 激活模式,需要让矩形跟随移动
                            if squareManager.drag_active:
                                # 更新方块
                                squareManager.updateSquare(between_finger_tip[0],between_finger_tip[1])
                                if(line_len>100):
                                    # 取消激活
                                    squareManager.drag_active =False
                                    squareManager.active_index = -1

                            elif (line_len<100) and (squareManager.checkOverlay(between_finger_tip[0],between_finger_tip[1]) != -1 )and( squareManager.drag_active  == False):
                                    # 激活
                                    squareManager.drag_active =True
                                    # 计算距离
                                    squareManager.setLen(between_finger_tip[0],between_finger_tip[1])

                # 显示方块,传入本实例,主要为了半透明的处理
                squareManager.display(self)
                
                # 显示距离
                cv2.putText(self.image, "Distance:"+str(rect_percent_text), (10, 120),cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)

                # 显示当前激活
                cv2.putText(self.image, "Active:"+( "None" if squareManager.active_index == -1 else str(squareManager.active_index)), (10, 170),cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)

                # 显示刷新率FPS
                cTime = time.time()
                fps_text = 1/(cTime-fpsTime)
                fpsTime = cTime
                cv2.putText(self.image, "FPS: " + str(int(fps_text)), (10, 70),
                            cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)
                # 显示画面
                # self.image = cv2.resize(self.image, (resize_w//2, resize_h//2))
                cv2.imshow('virtual drag and drop', self.image)
                
                if cv2.waitKey(5) & 0xFF == 27:
                    break
            cap.release()



# 开始程序
control = HandControlVolume()
control.recognize()


3.mediapipe实现手势控制音量大小(转载于恩培大佬)

效果如示:

源码如下:

"""
功能:手势操作电脑音量
1、使用OpenCV读取摄像头视频流;
2、识别手掌关键点像素坐标;
3、根据拇指和食指指尖的坐标,利用勾股定理计算距离;
4、将距离等比例转为音量大小,控制电脑音量
"
""


# 导入OpenCV
import cv2
# 导入mediapipe
import mediapipe as mp
# 导入电脑音量控制模块
from ctypes import cast, POINTER
from comtypes import CLSCTX_ALL
from pycaw.pycaw import AudioUtilities, IAudioEndpointVolume

# 导入其他依赖包
import time
import math
import numpy as np

class HandControlVolume:
    def __init__(self):
        # 初始化medialpipe
        self.mp_drawing = mp.solutions.drawing_utils
        self.mp_drawing_styles = mp.solutions.drawing_styles
        self.mp_hands = mp.solutions.hands

        # 获取电脑音量范围
        devices = AudioUtilities.GetSpeakers()
        interface = devices.Activate(
            IAudioEndpointVolume._iid_, CLSCTX_ALL, None)
        self.volume = cast(interface, POINTER(IAudioEndpointVolume))
        self.volume_range = self.volume.GetVolumeRange()

    # 主函数
    def recognize(self):
        # 计算刷新率
        fpsTime = time.time()

        # OpenCV读取视频流
        cap = cv2.VideoCapture(0)
        # 视频分辨率
        resize_w = 640
        resize_h = 480

        # 画面显示初始化参数
        rect_height = 0
        rect_percent_text = 0

        with self.mp_hands.Hands(min_detection_confidence=0.7,
                                 min_tracking_confidence=0.5,
                                 max_num_hands=2) as hands:
            while cap.isOpened():
                success, image = cap.read()
                image = cv2.resize(image, (resize_w, resize_h))

                if not success:
                    print("空帧.")
                    continue

                # 提高性能
                image.flags.writeable = False
                # 转为RGB
                image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
                # 镜像
                image = cv2.flip(image, 1)
                # mediapipe模型处理
                results = hands.process(image)

                image.flags.writeable = True
                image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
                # 判断是否有手掌
                if results.multi_hand_landmarks:
                    # 遍历每个手掌
                    for hand_landmarks in results.multi_hand_landmarks:
                        # 在画面标注手指
                        self.mp_drawing.draw_landmarks(
                            image,
                            hand_landmarks,
                            self.mp_hands.HAND_CONNECTIONS,
                            self.mp_drawing_styles.get_default_hand_landmarks_style(),
                            self.mp_drawing_styles.get_default_hand_connections_style())

                        # 解析手指,存入各个手指坐标
                        landmark_list = []
                        for landmark_id, finger_axis in enumerate(
                                hand_landmarks.landmark):
                            landmark_list.append([
                                landmark_id, finger_axis.x, finger_axis.y,
                                finger_axis.z
                            ])
                        if landmark_list:
                            # 获取大拇指指尖坐标
                            thumb_finger_tip = landmark_list[4]
                            thumb_finger_tip_x = math.ceil(thumb_finger_tip[1] * resize_w)
                            thumb_finger_tip_y = math.ceil(thumb_finger_tip[2] * resize_h)
                            # 获取食指指尖坐标
                            index_finger_tip = landmark_list[8]
                            index_finger_tip_x = math.ceil(index_finger_tip[1] * resize_w)
                            index_finger_tip_y = math.ceil(index_finger_tip[2] * resize_h)
                            # 中间点
                            finger_middle_point = (thumb_finger_tip_x+index_finger_tip_x)//2, (thumb_finger_tip_y+index_finger_tip_y)//2
                            # print(thumb_finger_tip_x)
                            thumb_finger_point = (thumb_finger_tip_x,thumb_finger_tip_y)
                            index_finger_point = (index_finger_tip_x,index_finger_tip_y)
                            # 画指尖2点
                            image = cv2.circle(image,thumb_finger_point,10,(255,0,255),-1)
                            image = cv2.circle(image,index_finger_point,10,(255,0,255),-1)
                            image = cv2.circle(image,finger_middle_point,10,(255,0,255),-1)
                            # 画2点连线
                            image = cv2.line(image,thumb_finger_point,index_finger_point,(255,0,255),5)
                            # 勾股定理计算长度
                            line_len = math.hypot((index_finger_tip_x-thumb_finger_tip_x),(index_finger_tip_y-thumb_finger_tip_y))

                            # 获取电脑最大最小音量
                            min_volume = self.volume_range[0]
                            max_volume = self.volume_range[1]
                            # 将指尖长度映射到音量上
                            vol = np.interp(line_len,[50,300],[min_volume,max_volume])
                            # 将指尖长度映射到矩形显示上
                            rect_height = np.interp(line_len,[50,300],[0,200])
                            rect_percent_text = np.interp(line_len,[50,300],[0,100])

                            # 设置电脑音量
                            self.volume.SetMasterVolumeLevel(vol, None)

                # 显示矩形
                cv2.putText(image, str(math.ceil(rect_percent_text))+"%", (10, 350),
                            cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)
                image = cv2.rectangle(image,(30,100),(70,300),(255, 0, 0),3)
                image = cv2.rectangle(image,(30,math.ceil(300-rect_height)),(70,300),(255, 0, 0),-1)

                # 显示刷新率FPS
                cTime = time.time()
                fps_text = 1/(cTime-fpsTime)
                fpsTime = cTime
                cv2.putText(image, "FPS: " + str(int(fps_text)), (10, 70),
                            cv2.FONT_HERSHEY_PLAIN, 3, (255, 0, 0), 3)
                # 显示画面
                cv2.imshow('MediaPipe Hands', image)
                if cv2.waitKey(5) & 0xFF == 27:
                    break
            cap.release()

# 开始程序
control = HandControlVolume()
control.recognize()

reference:

源码来源:恩培大佬




Python“宝藏级”公众号【Python之王】专注于Python领域,会爬虫,数分,C++,tensorflow和Pytorch等等

近 2年共原创 100+ 篇技术文章。创作的精品文章系列有:

日常收集整理了一批不错的 Python 学习资料,有需要的小伙可以自行免费领取。

获取方式如下:公众号回复资料领取Python等系列笔记,项目,书籍,直接套上模板就可以用了。资料包含算法、python、算法小抄、力扣刷题手册和 C++ 等学习资料!

浏览 126
点赞
评论
收藏
分享

手机扫一扫分享

分享
举报
评论
图片
表情
推荐
点赞
评论
收藏
分享

手机扫一扫分享

分享
举报