您好,我在编译中有此错误,任何人都可以帮帮我吗
##错误##
raceback(最近一次通话最后):文件“C:\Users\ASUS\Desktop\PFE Master Eye Tracker 2021\PFE_Eye_Tracker_code\opencv-eye-tracking-master\eye_control.py”,第 47 行,在学生帧 = cv2.equalizeHist( frame[y+(h*0.25):(y+h), x:(x+w)]) TypeError: slice indices must be integers or None or have an index method
进程以退出代码 -1073740791 (0xC0000409) 结束
##CODE##这是我的代码
import numpy as np
import cv2
import time
cap = cv2.VideoCapture(0) #initialize video capture
left_counter=0 #counter for left movement
right_counter=0 #counter for right movement
th_value=5 #changeable threshold value
def thresholding( value ): # function to threshold and give either left or right
global left_counter
global right_counter
if (value<=54): #check the parameter is less than equal or greater than range to
left_counter=left_counter+1 #increment left counter
if (left_counter>th_value): #if left counter is greater than threshold value
print ('RIGHT') #the eye is left
left_counter=0 #reset the counter
elif(value>=54): # same procedure for right eye
right_counter=right_counter+1
if(right_counter>th_value):
print ('LEFT')
right_counter=0
while 1:
ret, frame = cap.read()
cv2.line(frame, (320,0), (320,480), (0,200,0), 2)
cv2.line(frame, (0,200), (640,200), (0,200,0), 2)
if ret==True:
col=frame
frame = cv2.cvtColor(frame,cv2.COLOR_RGB2GRAY)
pupilFrame=frame
clahe=frame
blur=frame
edges=frame
eyes = cv2.CascadeClassifier('haarcascade_eye.xml')
detected = eyes.detectMultiScale(frame, 1.3, 5)
for (x,y,w,h) in detected: #similar to face detection
cv2.rectangle(frame, (x,y), ((x+w),(y+h)), (0,0,255),1) #draw rectangle around eyes
cv2.line(frame, (x,y), ((x+w,y+h)), (0,0,255),1) #draw cross
cv2.line(frame, (x+w,y), ((x,y+h)), (0,0,255),1)
# using histogram equalization of better image.
pupilFrame = cv2.equalizeHist(frame[y+(h*.25):(y+h), x:(x+w)])
cl1 = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8)) #set grid size
clahe = cl1.apply(pupilFrame) #clahe
blur = cv2.medianBlur(clahe, 7) #median blur
circles = cv2.HoughCircles(blur ,cv2.cv.CV_HOUGH_GRADIENT,1,20,param1=50,param2=30,minRadius=7,maxRadius=21) #houghcircles
if circles is not None: #if atleast 1 is detected
circles = np.round(circles[0, :]).astype("int") #change float to integer
print ('integer',circles)
for (x,y,r) in circles:
cv2.circle(pupilFrame, (x, y), r, (0, 255, 255), 2)
cv2.rectangle(pupilFrame, (x - 5, y - 5), (x + 5, y + 5), (0, 128, 255), -1)
#set thresholds
thresholding(x)
#frame = cv2.medianBlur(frame,5)
cv2.imshow('image',pupilFrame)
cv2.imshow('clahe', clahe)
cv2.imshow('blur', blur)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()