摘要:一般调高饱和度会降低中相对较低的数值,凸显主要颜色的纯度。对当前唯一的方法霍夫梯度法,它表示在检测阶段圆心的累加器阈值。第八个参数,类型的有默认值,表示圆半径的最小值。
目录
为什么用HSV空间而不是RGB空间?
因为RGB通道并不能很好地反映出物体具体的颜色信息。
而HSV空间能够非常直观的表达色彩的明暗、色调、以及鲜艳程度,方便进行颜色之间的对比。
(RGB受光线影响很大,所以采取HSV)
这里用HSV的目的:得到合适的二值图。
Hue(H):色调、色相(具体的颜色)
Saturation(S):饱和度、色彩纯净度
Value(V):明度
Hue范围是[0,179],饱和范围是[0,255],值范围是[0,255]
(写代码的时候,犯了蠢错误:把3个都设成了179,发现死活调不对)
Hue:色相(具体的颜色)
明度:色彩的明亮程度,单通道亮度(并不等同于整体发光量)。
(明度越高越白,越低越黑,一般提高明度会同时提高R、G、B三通道的数值)
Saturation:饱和度、色彩纯度(越低越灰,越高越纯) 。
(一般调高饱和度会降低RGB中相对较低的数值,凸显主要颜色的纯度。 )
B站视频讲解:
短动画慢语速1分钟讲清影视调色中色彩形成原理基础——RGB与HSV
这次的初始化,除了电机、舵机、窗口等等的初始化,还有滑动条的初始设置。
在创建滑动条之前:
1、需要设置窗口名称,不然窗口都没有,自然也就无法设置滑动条了。
2、创建回调函数。
def nothing(*arg): pass
def Trackbar_Init(): # 1 create windows cv2.namedWindow("h_binary") cv2.namedWindow("s_binary") cv2.namedWindow("v_binary")
# 2 Create Trackbar cv2.createTrackbar("hmin", "h_binary", 6, 179, nothing) cv2.createTrackbar("hmax", "h_binary", 26, 179, nothing) cv2.createTrackbar("smin", "s_binary", 110, 255, nothing) cv2.createTrackbar("smax", "s_binary", 255, 255, nothing) cv2.createTrackbar("vmin", "v_binary", 140, 255, nothing) cv2.createTrackbar("vmax", "v_binary", 255, 255, nothing) # 创建滑动条 滑动条值名称 窗口名称 滑动条值 滑动条阈值 回调函数
def Motor_Init(): global L_Motor, R_Motor L_Motor= GPIO.PWM(l_motor,100) R_Motor = GPIO.PWM(r_motor,100) L_Motor.start(0) R_Motor.start(0)def Direction_Init(): GPIO.setup(left_back,GPIO.OUT) GPIO.setup(left_front,GPIO.OUT) GPIO.setup(l_motor,GPIO.OUT) GPIO.setup(right_front,GPIO.OUT) GPIO.setup(right_back,GPIO.OUT) GPIO.setup(r_motor,GPIO.OUT)def Servo_Init(): global pwm_servo pwm_servo=Adafruit_PCA9685.PCA9685()def Trackbar_Init(): # 1 create windows cv2.namedWindow("h_binary") cv2.namedWindow("s_binary") cv2.namedWindow("v_binary") # 2 Create Trackbar cv2.createTrackbar("hmin", "h_binary", 6, 179, nothing) cv2.createTrackbar("hmax", "h_binary", 26, 179, nothing) cv2.createTrackbar("smin", "s_binary", 110, 255, nothing) cv2.createTrackbar("smax", "s_binary", 255, 255, nothing) cv2.createTrackbar("vmin", "v_binary", 140, 255, nothing) cv2.createTrackbar("vmax", "v_binary", 255, 255, nothing) # 创建滑动条 滑动条值名称 窗口名称 滑动条值 滑动条阈值 回调函数def Init(): GPIO.setwarnings(False) GPIO.setmode(GPIO.BCM) Direction_Init() Servo_Init() Motor_Init() Trackbar_Init()# 回调函数def nothing(*arg): pass
常规操作:向前、向后、向左、向右、停止。
def Front(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,1) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,1) #right_front GPIO.output(right_back,0) #right_back def Back(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,0) #left_front GPIO.output(left_back,1) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,0) #right_front GPIO.output(right_back,1) #right_backdef Left(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,0) #left_front GPIO.output(left_back,1) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,1) #right_front GPIO.output(right_back,0) #right_backdef Right(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,1) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,0) #right_front GPIO.output(right_back,1) #right_backdef Stop(): L_Motor.ChangeDutyCycle(0) GPIO.output(left_front,0) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(0) GPIO.output(right_front,0) #right_front GPIO.output(right_back,0) #right_back
def set_servo_angle(channel,angle): angle=4096*((angle*11)+500)/20000 pwm_servo.set_pwm_freq(50) #frequency==50Hz (servo) pwm_servo.set_pwm(channel,0,int(angle))
set_servo_angle(4, 110) #top servo lengthwise #0:back 180:front set_servo_angle(5, 90) #bottom servo crosswise #0:left 180:right
# 1 get trackbar"s value hmin = cv2.getTrackbarPos("hmin", "h_binary") hmax = cv2.getTrackbarPos("hmax", "h_binary") smin = cv2.getTrackbarPos("smin", "s_binary") smax = cv2.getTrackbarPos("smax", "s_binary") vmin = cv2.getTrackbarPos("vmin", "v_binary") vmax = cv2.getTrackbarPos("vmax", "v_binary")
# 2 to HSV hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) cv2.imshow("hsv", hsv) h, s, v = cv2.split(hsv)
分别对H、S、V三个道阈值处理:
# 3 set threshold (binary image) # if value in (min, max):white; otherwise:black h_binary = cv2.inRange(np.array(h), np.array(hmin), np.array(hmax)) s_binary = cv2.inRange(np.array(s), np.array(smin), np.array(smax)) v_binary = cv2.inRange(np.array(v), np.array(vmin), np.array(vmax))# 5 Show cv2.imshow("h_binary", h_binary) cv2.imshow("s_binary", s_binary) cv2.imshow("v_binary", v_binary)
对H、S、V三个通道与操作。(H&&S&&V)
# 4 get binary binary = cv2.bitwise_and(h_binary, cv2.bitwise_and(s_binary, v_binary))
# 在HSV色彩空间下得到二值图def Get_HSV(image): # 1 get trackbar"s value hmin = cv2.getTrackbarPos("hmin", "h_binary") hmax = cv2.getTrackbarPos("hmax", "h_binary") smin = cv2.getTrackbarPos("smin", "s_binary") smax = cv2.getTrackbarPos("smax", "s_binary") vmin = cv2.getTrackbarPos("vmin", "v_binary") vmax = cv2.getTrackbarPos("vmax", "v_binary") # 2 to HSV hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) cv2.imshow("hsv", hsv) h, s, v = cv2.split(hsv) # 3 set threshold (binary image) # if value in (min, max):white; otherwise:black h_binary = cv2.inRange(np.array(h), np.array(hmin), np.array(hmax)) s_binary = cv2.inRange(np.array(s), np.array(smin), np.array(smax)) v_binary = cv2.inRange(np.array(v), np.array(vmin), np.array(vmax)) # 4 get binary(对H、S、V三个通道分别与操作) binary = cv2.bitwise_and(h_binary, cv2.bitwise_and(s_binary, v_binary)) # 5 Show cv2.imshow("h_binary", h_binary) cv2.imshow("s_binary", s_binary) cv2.imshow("v_binary", v_binary) cv2.imshow("binary", binary) return binary
# 1 Capture the frames ret, frame = camera.read() image = frame cv2.imshow("frame", frame)
(Get_HSV(frame)其实就是步骤四的HSV处理)
# 2 get HSV binary = Get_HSV(frame)
# 3 Gausi blur blur = cv2.GaussianBlur(binary,(9,9),0)
# 4 Open kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9,9)) Open = cv2.morphologyEx(blur, cv2.MORPH_OPEN, kernel) cv2.imshow("Open",Open)
# 5 Close Close = cv2.morphologyEx(Open, cv2.MORPH_CLOSE, kernel) cv2.imshow("Close",Close)
霍夫变换圆检测是基于图像梯度实现:
圆心检测的原理︰圆心是圆周法线的交汇处,设置一个阈值,在某点的相交的直线的条数大于这个阈值就认为该交汇点为圆心。
圆半径确定原理:圆心到圆周上的距离〔半径)是相同的,设置一个阈值,只要相同距离的数量大于该阈值,就认为该距离是该圆心的半径。
API
def HoughCircles(image: Any, method: Any, dp: Any, minDist: Any, circles: Any = None, param1: Any = None, param2: Any = None, minRadius: Any = None, maxRadius: Any = None) -> None
参数:
- 第二个参数,int类型的method,即使用的检测方法,目前OpenCV中就霍夫梯度法一种可以使用,它的标识符为CV_HOUGH_GRADIENT,在此参数处填这个标识符即可。
- 第三个参数,double类型的dp,用来检测圆心的累加器图像的分辨率于输入图像之比的倒数,且此参数允许创建一个比输入图像分辨率低的累加器。上述文字不好理解的话,来看例子吧。例如,如果dp= 1时,累加器和输入图像具有相同的分辨率。如果dp=2,累加器便有输入图像一半那么大的宽度和高度。
- 第四个参数,double类型的minDist,为霍夫变换检测到的圆的圆心之间的最小距离,即让我们的算法能明显区分的两个不同圆之间的最小距离。这个参数如果太小的话,多个相邻的圆可能被错误地检测成了一个重合的圆。反之,这个参数设置太大的话,某些圆就不能被检测出来了。
- 第五个参数,InputArray类型的circles,经过调用HoughCircles函数后此参数存储了检测到的圆的输出矢量,每个矢量由包含了3个元素的浮点矢量(x, y, radius)表示。
- 第六个参数,double类型的param1,有默认值100。它是第三个参数method设置的检测方法的对应的参数。对当前唯一的方法霍夫梯度法CV_HOUGH_GRADIENT,它表示传递给canny边缘检测算子的高阈值,而低阈值为高阈值的一半。
- 第七个参数,double类型的param2,也有默认值100。它是第三个参数method设置的检测方法的对应的参数。对当前唯一的方法霍夫梯度法CV_HOUGH_GRADIENT,它表示在检测阶段圆心的累加器阈值。它越小的话,就可以检测到更多根本不存在的圆,而它越大的话,能通过检测的圆就更加接近完美的圆形了。
- 第八个参数,int类型的minRadius,有默认值0,表示圆半径的最小值。
- 第九个参数,int类型的maxRadius,也有默认值0,表示圆半径的最大值。
# 6 Hough Circle detect circles = cv2.HoughCircles(Close,cv2.HOUGH_GRADIENT,2,120,param1=120,param2=20,minRadius=20,maxRadius=0) # param2:决定圆能否被检测到(越少越容易检测到圆,但相应的也更容易出错)
# 1 获取圆的圆心和半径 x, y, r = int(circles[0][0][0]),int(circles[0][0][1]),int(circles[0][0][2]) print(x, y, r)
# 2 画圆 cv2.circle(image, (x, y), r, (255,0,255),5) cv2.imshow("image", image)
# 图像处理def Image_Processing(): global h, s, v # 1 Capture the frames ret, frame = camera.read() image = frame cv2.imshow("frame", frame) # 2 get HSV binary = Get_HSV(frame) # 3 Gausi blur blur = cv2.GaussianBlur(binary,(9,9),0) # 4 Open kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9,9)) Open = cv2.morphologyEx(blur, cv2.MORPH_OPEN, kernel) cv2.imshow("Open",Open) # 5 Close Close = cv2.morphologyEx(Open, cv2.MORPH_CLOSE, kernel) cv2.imshow("Close",Close) # 6 Hough Circle detect circles = cv2.HoughCircles(Close,cv2.HOUGH_GRADIENT,2,120,param1=120,param2=20,minRadius=20,maxRadius=0) # param2:决定圆能否被检测到(越少越容易检测到圆,但相应的也更容易出错) # judge if circles is exist if circles is not None: # 1 获取圆的圆心和半径 x, y, r = int(circles[0][0][0]),int(circles[0][0][1]),int(circles[0][0][2]) print(x, y, r) # 2 画圆 cv2.circle(image, (x, y), r, (255,0,255),5) cv2.imshow("image", image) else: (x,y),r = (0,0), 0 return (x,y), r
根据检测到的圆,获取到的坐标、半径,进行运动控制。
这里可以做到跟踪小球,前进和后退相配合,“敌进我退,敌退我进”。
# 运动控制(这里可以做到跟踪小球,前景和后退相配合,“敌进我退,敌退我进”)def Move((x,y), r): low_xlimit = width/4 high_xlimit = 0.75 * width #low_ylimit = 3/4 * height ylimit = 0.75 * height print(high_xlimit, ylimit) # 没检测到,停止不动 if x==0: Stop() # 检测到在图片0.75以上的区域(距离正常) elif x>low_xlimit and xlow_xlimit and x=ylimit: Back(60) # 在左0.25区域,向左跟踪 elif xhigh_xlimit: Right(60)
#Ball Tracking(HSV)import RPi.GPIO as GPIOimport timeimport Adafruit_PCA9685import numpy as npimport cv2#set capture windowwidth, height = 320, 240camera = cv2.VideoCapture(0)camera.set(3,width) camera.set(4,height) l_motor = 18left_front = 22left_back = 27r_motor = 23right_front = 25right_back = 24def Motor_Init(): global L_Motor, R_Motor L_Motor= GPIO.PWM(l_motor,100) R_Motor = GPIO.PWM(r_motor,100) L_Motor.start(0) R_Motor.start(0)def Direction_Init(): GPIO.setup(left_back,GPIO.OUT) GPIO.setup(left_front,GPIO.OUT) GPIO.setup(l_motor,GPIO.OUT) GPIO.setup(right_front,GPIO.OUT) GPIO.setup(right_back,GPIO.OUT) GPIO.setup(r_motor,GPIO.OUT)def Servo_Init(): global pwm_servo pwm_servo=Adafruit_PCA9685.PCA9685()def Trackbar_Init(): # 1 create windows cv2.namedWindow("h_binary") cv2.namedWindow("s_binary") cv2.namedWindow("v_binary") # 2 Create Trackbar cv2.createTrackbar("hmin", "h_binary", 6, 179, nothing) cv2.createTrackbar("hmax", "h_binary", 26, 179, nothing) cv2.createTrackbar("smin", "s_binary", 110, 255, nothing) cv2.createTrackbar("smax", "s_binary", 255, 255, nothing) cv2.createTrackbar("vmin", "v_binary", 140, 255, nothing) cv2.createTrackbar("vmax", "v_binary", 255, 255, nothing) # 创建滑动条 滑动条值名称 窗口名称 滑动条值 滑动条阈值 回调函数def Init(): GPIO.setwarnings(False) GPIO.setmode(GPIO.BCM) Direction_Init() Servo_Init() Motor_Init() Trackbar_Init()def Front(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,1) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,1) #right_front GPIO.output(right_back,0) #right_back def Back(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,0) #left_front GPIO.output(left_back,1) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,0) #right_front GPIO.output(right_back,1) #right_backdef Left(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,0) #left_front GPIO.output(left_back,1) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,1) #right_front GPIO.output(right_back,0) #right_backdef Right(speed): L_Motor.ChangeDutyCycle(speed) GPIO.output(left_front,1) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(speed) GPIO.output(right_front,0) #right_front GPIO.output(right_back,1) #right_backdef Stop(): L_Motor.ChangeDutyCycle(0) GPIO.output(left_front,0) #left_front GPIO.output(left_back,0) #left_back R_Motor.ChangeDutyCycle(0) GPIO.output(right_front,0) #right_front GPIO.output(right_back,0) #right_backdef set_servo_angle(channel,angle): angle=4096*((angle*11)+500)/20000 pwm_servo.set_pwm_freq(50) #frequency==50Hz (servo) pwm_servo.set_pwm(channel,0,int(angle))# 回调函数def nothing(*arg): pass# 在HSV色彩空间下得到二值图def Get_HSV(image): # 1 get trackbar"s value hmin = cv2.getTrackbarPos("hmin", "h_binary") hmax = cv2.getTrackbarPos("hmax", "h_binary") smin = cv2.getTrackbarPos("smin", "s_binary") smax = cv2.getTrackbarPos("smax", "s_binary") vmin = cv2.getTrackbarPos("vmin", "v_binary") vmax = cv2.getTrackbarPos("vmax", "v_binary") # 2 to HSV hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) cv2.imshow("hsv", hsv) h, s, v = cv2.split(hsv) # 3 set threshold (binary image) # if value in (min, max):white; otherwise:black h_binary = cv2.inRange(np.array(h), np.array(hmin), np.array(hmax)) s_binary = cv2.inRange(np.array(s), np.array(smin), np.array(smax)) v_binary = cv2.inRange(np.array(v), np.array(vmin), np.array(vmax)) # 4 get binary(对H、S、V三个通道分别与操作) binary = cv2.bitwise_and(h_binary, cv2.bitwise_and(s_binary, v_binary)) # 5 Show cv2.imshow("h_binary", h_binary) cv2.imshow("s_binary", s_binary) cv2.imshow("v_binary", v_binary) cv2.imshow("binary", binary) return binary# 图像处理def Image_Processing(): global h, s, v # 1 Capture the frames ret, frame = camera.read() image = frame cv2.imshow("frame", frame) # 2 get HSV binary = Get_HSV(frame) # 3 Gausi blur blur = cv2.GaussianBlur(binary,(9,9),0) # 4 Open kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9,9)) Open = cv2.morphologyEx(blur, cv2.MORPH_OPEN, kernel) cv2.imshow("Open",Open) # 5 Close Close = cv2.morphologyEx(Open, cv2.MORPH_CLOSE, kernel) cv2.imshow("Close",Close) # 6 Hough Circle detect circles = cv2.HoughCircles(Close,cv2.HOUGH_GRADIENT,2,120,param1=120,param2=20,minRadius=20,maxRadius=0) # param2:决定圆能否被检测到(越少越容易检测到圆,但相应的也更容易出错) # judge if circles is exist if circles is not None: # 1 获取圆的圆心和半径 x, y, r = int(circles[0][0][0]),int(circles[0][0][1]),int(circles[0][0][2]) print(x, y, r) # 2 画圆 cv2.circle(image, (x, y), r, (255,0,255),5) cv2.imshow("image", image) else: (x,y),r = (0,0), 0 return (x,y), r# 运动控制(这里可以做到跟踪小球,前景和后退相配合,“敌进我退,敌退我进”)def Move((x,y), r): low_xlimit = width/4 high_xlimit = 0.75 * width #low_ylimit = 3/4 * height ylimit = 0.75 * height print(high_xlimit, ylimit) # 没检测到,停止不动 if x==0: Stop() # 检测到在图片0.75以上的区域(距离正常) elif x>low_xlimit and xlow_xlimit and x=ylimit: Back(60) # 在左0.25区域,向左跟踪 elif xhigh_xlimit: Right(60) if __name__ == "__main__": Init() set_servo_angle(4, 110) #top servo lengthwise #0:back 180:front set_servo_angle(5, 90) #bottom servo crosswise #0:left 180:right while 1: # 1 Image Process (x,y), r = Image_Processing() # 2 Move Move((x,y), r) # must include this codes(otherwise you can"t open camera successfully) if cv2.waitKey(1) & 0xFF == ord("q"): Stop() GPIO.cleanup() break
这里的HSV是根据我自己当前的情况调节的,更改场景以后可能需要重新调节H、S、V三通道的阈值(max && min)
基础的视觉检测+运动,没有太多的运动控制算法(PID等等都没有涉及到)。有好的想法和建议欢迎交流讨论,Thanks♪(・ω・)ノ
文章版权归作者所有,未经允许请勿转载,若此文章存在违规行为,您可以联系管理员删除。
转载请注明本文地址:https://www.ucloud.cn/yun/119986.html
摘要:在我已经制作完成一辆可以运行的遥控车时,公司发布了一个自驾车项目,来展示自动驾驶汽车的工作原理。需要注意的是,这里用的都是语言而非,其主要原因有两个一方面,近来似乎已成为运用机器学习技术时实际使用的语言。 最近,Mapbox 的 Android 工程师 Antonio 使用计算机视觉和机器学习技术,为他的女儿 Violeta 重新制作了一台遥控车。接下来我们看看 Antonio 是如何...
阅读 3290·2023-04-26 02:40
阅读 4641·2021-09-22 15:22
阅读 1577·2021-09-22 10:02
阅读 3477·2021-08-11 10:23
阅读 1390·2019-08-30 15:55
阅读 2488·2019-08-30 12:48
阅读 586·2019-08-30 11:04
阅读 698·2019-08-29 16:29