Background: 2022 Smart Car Competition Baidu Improvement Group
Idea: first take the track and adjust the threshold through HSV, then get the binarized image, and perform line inspection on the binarized image;
The idea of patrolling the line: start from the center of the last line of the picture and sweep the line to the left and right sides: sweep the left line and sweep the right line;
Take the left line as an example: (The picture size is 480*640)
The picture starts from the last row and loops to the first row as the outer loop (set as i), and starts from the middle line and subtracts one to the left as the inner loop (set as j); record the jump point of the row: that is, if the column of the row If it is white (255), the next column is black (0), record its column label (j). If this condition is not met, it means that there is no jump point, that is, the row is wireless, and it is recorded as 0; record it in an array (append Methods).
The idea is the same as the line on the right, except that if there is no line, it will be recorded as the largest column label on the right (639)
Track element processing:
The track has crosses, sharp turns, double crosses, and missing corners and tracks due to the location of the camera installation
Acute angle:
：
Double Cross:
Cross:
Notch, etc.:
...will not show them one by one.
Processing ideas: 1. Acute angle elements:
Because of the acute angle, the line sweeping method is not applicable; however, the acute angle will only appear on the top of the picture, either the left acute angle or the right acute angle. Our track only has the left acute angle, and the proportion is small, so directly judge the acute angle, and then do not Sweep lines at acute angles;
Effect:
Code: If the leftmost line of the current line is a non-lane (0), the next line (up) is a lane (non-0), and the right line is a non-lane, the loop is interrupted; because acute angles only appear in left turns, we track The crosses appear when turning left, so there will be no missed sweeps (if you turn right at the crosses, there may be missed sweeps).
2. Cross and notch:
Idea: first calculate the inflection point, and then judge the supplementary line through the inflection point; the supplementary line directly connects the upper and lower inflection points: oblique intercept equation.
Calculation of the inflection point: Calculate the jump point of the left (right) side line: from 0 (black) to non-0 (white) is the left (right) upper inflection point, from non-0 (white) to 0 (black) is the left (Right) Lower inflection point. (Use four variables to record whether there is an inflection point, and if so, use four variables to store the coordinates of the inflection point)
A. Complement the inflection point first, that is, if there is only one inflection point on the left (right) side, the other inflection point will be assigned by this inflection point. See the program for details: if there is only an upper inflection point, record the lower inflection point as the point in the same column at the bottom row...
B. Judging how to make up the line according to the inflection point mark
Compensation line judgment: as long as there is an inflection point, the line will be supplemented
Effect:
The position of the inflection point can be adjusted by the variable sm in the code
Code: (It is time-consuming to display the photos in the function, and the return value is the midline array of Xunde)
import cv2
import numpy
import numpy as np
def nothing(*arg):
pass
icol = (18, 0, 196, 36, 255, 255)
#path = "test/cruise/"
# Show the original image.
#frame = path+str(961)+'.jpg'
#frame = cv2.imread(frame)
l = [16, 45, 65] # [17, 55, 128]#阈值
h = [44, 255, 255] # [24, 255, 255]#阈值
def zh_ch(string):
return string.encode("gbk").decode('UTF-8', errors='ignore')
# mm = 320
cap = cv2.VideoCapture(0)
def XunX(img,SX):
cv2.imshow('frame', img)
# Blur methods available, comment or uncomment to try different blur methods.
frameBGR = cv2.GaussianBlur(img, (7, 7), 0)
hsv = cv2.cvtColor(frameBGR, cv2.COLOR_BGR2HSV)
# HSV values to define a colour range.
colorLow = numpy.array([16, 45, 65] ) # [17, 55, 128]
colorHigh = numpy.array([44, 225, 225]) # [24, 225, 225]
mask = cv2.inRange(hsv, colorLow, colorHigh)
# Show the first mask
cv2.imshow('mask-plain', mask)
kernal = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7, 7))
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernal)
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernal)
cv2.imshow('mask-plain', mask)
# Find_Line(mask)
left = np.array([])
right = np.array([])
leftb = np.array([])
rightb = np.array([])
medim = np.array([])
# len = []#记录边线丢失
le = [[]]
ri = [[]]
zz = 0 # 记录左缺陷时上一个坐标
yy = 639
SX = SX # 开始扫线时的y坐标
SXp = np.array([])
SS = 0
left_up = left_down = right_up = right_down = 0
for i in range(480, 1, -1):
# 扫左线
for j in range(SX + 1, 0, -1):
# 中线：319
if (mask[i - 1][j] == 0) and (mask[i - 1][j - 1] == 255):
# 记录左跳变点的列值
left = np.append(left, j)
leftb = np.append(leftb, j)
zz = j
break
# 现在缺线
elif (j == 1):
lk = 0
left = np.append(left, lk)
leftb = np.append(leftb, lk)
# 扫右线
for j1 in range(SX + 2, 639, 1):
if (mask[i - 1][j1] == 0) and (mask[i - 1][j1 + 1] == 255):
# 记录右跳变点的列值
right = np.append(right, j1)
rightb = np.append(rightb, j1)
yy = j1
break
# 缺线
elif (j1 == 638):
lk = 639
rightb = np.append(rightb, lk)
right = np.append(right, lk)
if (left[480 - i] == 0 and right[480 - i] != 639 and mask[i - 2][0] != 0 ):
break
# if (right[480 - i] == 639 and left[480 - i] != 0 and mask[i - 2][639] != 255 ):
# break
SX = int((left[480 - i] + right[480 - i]) / 2)
SXp = np.append(SXp, SX)
# 找拐点
sm = 30
for i in range(len(left) - 10):
if (left[i] == 0 and left[i + 3] == 0 and left[i + 5] > 0 and left[i + 10] > 0):
left_up = 1
left_up1 = (i + min(len(left[i+2:]),sm), left[i + min(len(left[i+2:]),sm)]) # 480 - i-sm
if (left[i] > 0 and left[i + 3] > 0 and left[i + 5] == 0 and left[i + 10] == 0):
left_down = 1
left_down1 = (i - min(i,sm), left[i - min(i,sm)]) # 480 - i+sm
if (right[i] == 639 and right[i + 3] == 639 and right[i + 5] < 639 and right[i + 10] <= 639):
right_up = 1
right_up1 = (i + min(len(left[i+2:]),sm), right[i + min(len(left[i+2:]),sm)])
if (right[i] < 639 and right[i + 3] < 639 and right[i + 5] == 639 and right[i + 10] == 639):
right_down = 1
right_down1 = (i - min(i,sm), right[i - min(i,sm)]) # -1
# 判断元素：补线操作
print(left_up, left_down, right_up, right_down)
if (left_up and not left_down):
left_down1 = (0, left_up1[1])
elif (not left_up and left_down):
left_up1 = (479, left_down1[1])
if (right_up and not right_down):
right_down1 = (0, right_up1[1])
elif (not right_up and right_down):
right_up1 = (479, right_down1[1])
# 左右的上下拐点同时出现
#if (left_up and right_up) or (left_down and right_down):
if (left_up or right_up) or (left_down or right_down):
# if(right_up1[0] - left_up1[0] <= 20):
for k in range(len(leftb)):
if (k >= left_down1[0] and k <= left_up1[0]) or (k <= left_down1[0] and k >= left_up1[0]):
leftb[k] = ((left_up1[1] - left_down1[1]) / (left_up1[0] -
left_down1[0])) * (k - left_up1[0]) + left_up1[1]
if (k >= right_down1[0] and k <= right_up1[0]) or (k <= right_down1[0] and k >= right_up1[0]):
rightb[k] = ((right_up1[1] - right_down1[1]) / (right_up1[0] -
right_down1[0])) * (k - right_up1[0]) + right_up1[1]
# # 同时出现左上下，或右上下：
# if (left_up and left_down) or (right_up and right_down):
# # if(right_up1[0] - left_up1[0] <= 20):
# for k in range(len(leftb)):
# if (k >= left_down1[0] and k <= left_up1[0]) or (k <= left_down1[0] and k >= left_up1[0]):
# leftb[k] = ((left_up1[1] - left_down1[1]) / (left_up1[0] -
# left_down1[0])) * (k - left_up1[0]) + left_up1[1]
# if (k >= right_down1[0] and k <= right_up1[0]) or (k <= right_down1[0] and k >= right_up1[0]):
# rightb[k] = ((right_up1[1] - right_down1[1]) / (right_up1[0] -
# right_down1[0])) * (k - right_up1[0]) + right_up1[1]
# left_up = left_down = right_up = right_down = 0
# if
medim = (leftb + rightb) / 2
# img = [mask,mask3,mask4,mask5]
pl = ['left', 'right', 'l&r', 'no']
o = 0
img = mask
cg = len(left)-1
print(cg,len(leftb),len(right))
for k in range(cg, -1, -1):
point = (int(medim[k]), 479 - k) # 左
# point2 = (SXp[k], 478-k)
point3 = (int(leftb[k]), 479 - k) # 中
point1 = (int(rightb[k]), 479 - k) # 右
cv2.circle(img, point, 1, (255, 0, 255), 0)
cv2.circle(img, point1, 1, (255, 0, 255), 0)
cv2.circle(img, point3, 1, (255, 0, 255), 0)
cv2.imshow(pl[o], img)
cv2.waitKey(100000)
return SX,(leftb + rightb) / 2,SXp
mm = 320#扫线开始的坐标
cap = cv2.VideoCapture(0)
while True:
test, frame = cap.read()
print(frame.shape)
mm,XB,SXp = XunX(frame, mm)
cv2.destroyAllWindows()
Effect:
Appendix (threshold debugging code and article):
https://blog.csdn.net/qq_35831978/article/details/106988028?spm=1001.2014.3001.5506
Continually updated