当我们把手近似出来后会得到一组轮廓的点坐标，我自己手势识别的思路就是根据点坐标来判断手势。

根据坐标识别

寻找最低点

所谓寻找最低点，就是寻找手掌最下面点的坐标

比如这幅图，最左下角的点就是手掌的最低点，获得的方法是使用numpy的np.argmax函数，将维度设置为图片高的维度，获得最大的高就是获得最低的点。

maxindex = np.argmax(allpoint)

计算其他点与最低点的距离

获取成功之后就将最低点与其他点的距离计算出来：

#pt1点一
#pt2点二
def get_distance(pt1, pt2):
    distance = ((pt2[0] - pt1[0]) ** 2 + (pt2[1] - pt1[1]) ** 2) ** (0.5)
    return distance

for point in allpoint:
    distance = self.get_distance(self.allpoint[maxindex [1], :], point)
    if distance > 300:
        cv2.line(img,all_point[maxindex [1],:],point,bgr,4,16)

通过距离阈值判断手指根数和手势

然后通过这个距离阈值排除两根手指中间的点:

可以发现左边的图片线的数量比右边少了几根双指中间的。
最后我们只需要通过手指线的根数来判断手势就行了。

效果展现

完整代码

import cv2
import numpy as np
import random

class Img_handle:
    def __init__(self,area_min,lenth_min,distance,drawtype=0,area_max=100000,lenth_max=100000):
        #default area_min=20000 lenth_min=1000 distance=300
			#writer:csdn月明Mo
			#area_min 轮廓面积最小值
			#area_max 轮廓面积最大值
			#lenth_min 轮廓周长最小值
			#lenth_max 轮廓周长最大值
			#distance 距离阈值
			#drawtype 手势描绘类型，0为近似描绘 1为直线描绘
        self.area_min=area_min
        self.area_max=area_max
        self.lenth_min=lenth_min
        self.lenth_max=lenth_max
        self.distance=distance
        self.allpoint = []
        self.highHSV = np.array([15, 255, 255])
        self.lowHSV = np.array([0, 50, 50])
        self.drawtype=drawtype
	
	#writer:csdn月明Mo
	#Hsv阈值改变函数
    def change_Hsv(self,lowHSV,highHSV):
        self.highHSV=highHSV
        self.lowHSV=lowHSV

	#writer:csdn月明Mo
	#图片不失真resize函数
    def resize_img(self,size=list,img=None):
        size = [size[1], size[0], size[2]]
        mask = np.zeros(size, dtype=np.uint8)
        h, w = self.img.shape[0:2]
        dwh = min([size[0] / h, size[1] / w])
        self.img = cv2.resize(self.img, None, fx=dwh, fy=dwh)
        if h > w:
            dxy = int((size[1] - self.img.shape[1]) / 2)
            mask[:, dxy:self.img.shape[1] + dxy, :] = self.img
        else:
            dxy = int((size[0] - self.img.shape[0]) / 2)
            mask[dxy:self.img.shape[0] + dxy, :, :] = self.img
        return mask

	#writer:csdn月明Mo
	#图片处理函数
    def img_handle(self):

        self.img = cv2.cvtColor(self.img, cv2.COLOR_BGR2HSV)

        cv2.GaussianBlur(self.img, [5, 5], 0)

        self.img = cv2.inRange(self.img, self.lowHSV, self.highHSV)

        kernel = np.ones([3, 3], dtype=np.uint8)
        self.img = cv2.morphologyEx(self.img, cv2.MORPH_CLOSE, kernel, iterations=1)
        kernel = np.ones([3, 3], dtype=np.uint8)
        self.img = cv2.morphologyEx(self.img, cv2.MORPH_DILATE, kernel, iterations=1)

        contours, num = cv2.findContours(self.img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
        for contour in contours:
            area = cv2.contourArea(contour)
            lenth = cv2.arcLength(contour, True)
            if self.area_max >area > self.area_min and self.lenth_max >lenth > self.lenth_min:

                epsilon = 0.02 * cv2.arcLength(contour, True)
                self.allpoint = cv2.approxPolyDP(contour, epsilon, True)

                self.allpoint = self.allpoint.reshape(len(self.allpoint), 2)
                self.allpoint = np.array(self.allpoint, dtype=np.int32)
                if self.drawtype==0:

                    b = random.randint(0, 255)
                    g = random.randint(0, 255)
                    r = random.randint(0, 255)

                    cv2.polylines(self.output_img, [self.allpoint], True, [b, g, r], 4, 16)
                    
	#writer:csdn月明Mo
	#距离计算函数
    def get_distance(self, pt1, pt2):
        distance = ((pt2[0] - pt1[0]) ** 2 + (pt2[1] - pt1[1]) ** 2) ** (0.5)
        return distance

	#writer:csdn月明Mo
	#手势判断函数
    def detect(self):
        num = 0
        if np.any(self.allpoint):
            maxindex  = np.argmax(self.allpoint, axis=0)
            for point in self.allpoint:
                distance = self.get_distance(self.allpoint[maxindex [1], :], point)
                if distance > self.distance:
                    if self.drawtype ==1:

                        b = random.randint(0, 255)
                        g = random.randint(0, 255)
                        r = random.randint(0, 255)

                        cv2.line(self.output_img,self.allpoint[maxindex [1],:],point,[b, g, r],4,16)
                    num += 1
            if num == 1:
                cv2.putText(self.output_img, 'one', [10, 50], cv2.FONT_HERSHEY_SIMPLEX, 2, [0, 0, 255], thickness=4)
            elif num == 2:
                cv2.putText(self.output_img, 'two', [10, 50], cv2.FONT_HERSHEY_SIMPLEX, 2, [0, 0, 255], thickness=4)
            elif num == 3:
                cv2.putText(self.output_img, 'there', [10, 50], cv2.FONT_HERSHEY_SIMPLEX, 2, [0, 0, 255], thickness=4)
            elif num == 4:
                cv2.putText(self.output_img, 'four', [10, 50], cv2.FONT_HERSHEY_SIMPLEX, 2, [0, 0, 255], thickness=4)
            elif num == 5:
                cv2.putText(self.output_img, 'five', [10, 50], cv2.FONT_HERSHEY_SIMPLEX, 2, [0, 0, 255], thickness=4)

	#writer:csdn月明Mo
	#获取手势函数
    def get_hand(self,img):
        self.img = img
        if self.img.shape[0] != 480 and self.img.shape[1] != 640:
            self.img = self.resize_img([640,480,3])
        self.output_img=np.copy(self.img)
        self.img_handle()
        self.detect()
        return self.output_img


def main():
    video=cv2.VideoCapture(0)
    hand=Img_handle(20000,1000,280)
    while video.isOpened():
        res,img=video.read()
        if res== True:
            newimg=hand.get_hand(img)
            cv2.imshow('frams1', newimg)
        if cv2.waitKey(1)==ord('q'):
            break

    cv2.destroyAllWindows()
    video.release()

if __name__=='__main__':
    main()