opencv做颜色识别（python）

使用Raspberry摄像头抓图然后进行颜色识别。

#!/usr/bin/python

# -*- coding: UTF-8 -*-


import sys
import time


from picamera.array import PiRGBArray
from picamera import PiCamera
import numpy as np
import cv2


#颜色HSV范围
red_min = np.array([0, 128, 46])
red_max = np.array([5, 255,  255])
red2_min = np.array([156, 128,  46])
red2_max = np.array([180, 255,  255])


green_min = np.array([35, 128, 46])
green_max = np.array([77, 255, 255])


blue_min = np.array([100, 128, 46]) 
blue_max = np.array([124, 255, 255]) 


yellow_min = np.array([15, 128, 46])
yellow_max = np.array([34, 255, 255])


black_min = np.array([0,  0,  0]) 
black_max = np.array([180, 255, 10])

white_min = np.array([0, 0, 70]) 
white_max = np.array([180, 30, 255]) 

COLOR_ARRAY = [ [ red_min, red_max, 'red'],  [ red2_min, red2_max, 'red'],  [ green_min, green_max, 'green'], [ blue_min, blue_max, 'blue'],
                [yellow_min, yellow_max, 'yellow']  ]  #, [ black_min, black_max, 'black'],  [ white_min, white_max, 'white']  ]

camera = PiCamera()
camera.resolution = (640, 480)
camera.framerate = 25
rawCapture = PiRGBArray(camera, size=(640, 480))
time.sleep(0.1)


#while True:
for frame in camera.capture_continuous(rawCapture, format="bgr", use_video_port=True):
    frame = frame.array
    cv2.imwrite("frame.jpg", frame)
    #frame = cv2.imread("frame.jpg")
    hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
    cv2.imwrite("hsv.jpg", hsv)


    for (color_min, color_max, name)  in COLOR_ARRAY:
        mask=cv2.inRange(hsv,  color_min,  color_max)
        res=cv2.bitwise_and(frame, frame, mask=mask)
        #cv2.imshow("res",res)
        cv2.imwrite("2.jpg", res)


    #前面是为了得到一张二值图
        img = cv2.imread("2.jpg")
        h, w = img.shape[:2]


        blured = cv2.blur(img,(5,5))
        cv2.imwrite("blured.jpg", blured)
        ret, bright = cv2.threshold(blured,10,255,cv2.THRESH_BINARY)
    
        gray = cv2.cvtColor(bright,cv2.COLOR_BGR2GRAY)
        cv2.imwrite("gray.jpg", gray)
        
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (50, 50))
        opened = cv2.morphologyEx(gray, cv2.MORPH_OPEN, kernel)
        cv2.imwrite("opened.jpg", opened)
        closed = cv2.morphologyEx(opened, cv2.MORPH_CLOSE, kernel)
        #cv2.imshow("closed", closed)
        cv2.imwrite("closed.jpg", closed)


        contours, hierarchy = cv2.findContours(closed,cv2.RETR_LIST,cv2.CHAIN_APPROX_NONE) 
        cv2.drawContours(img,contours,-1,(0,0,255),3)
        cv2.imwrite("result.jpg",  img )
    
        #输出轮廓个数
        number = len(contours)
        print('Total:', number)
        if number  >=1:
            total = 0
            for i in range(0, number):
                total = total + len(contours[i])
                print 'NO:',i,' size:',  len(contours[i])
            if total > 400:
                print 'Currrent color is ', name
                cv2.destroyAllWindows()
                sys.exit()
            
    rawCapture.truncate(0)

难点是颜色范围的选择，可以使用取色工具对图片进行HSV采样，然后转换成opencv中的hsv（H/2, S*255/100, V*255/100）。

目前 识别饱和度比较高的颜色（如：红、蓝、绿、黄） 效果好，但对于黑白色不大容易检测，需要针对实际场景进行调试。