1

I wanted to create a model which can detect penalty area field lines. I managed to get desirable results for one specific image by changing some hyperparamaters, but the same code doesn't work for another image. i.e. the code is quite specific for an image. If I want to the code to work for every image what changes should I make.

One of the input images I took as input is here the output for the image is here One of the failure case which I got for an image while executing the code is here

this code is specific for the image provided above and is not working for some other image. Please let me know how can I improve my code.

import numpy as np
import cv2
import matplotlib.pyplot as plt
img = cv2.imread("C:/Users/user/Desktop/football1.jpg")

low_threshold = np.array([0,90,0], dtype=np.uint8)
high_threshold = np.array([170,255,255], dtype=np.uint8)  


mask1 = cv2.inRange(img, low_threshold, high_threshold)
cv2.imshow("mask1",mask1)

l_threshold = np.array([0, 115, 0], dtype=np.uint8)
h_threshold = np.array([75, 147, 172], dtype=np.uint8)  

mask2=cv2.inRange(img, low_threshold , high_threshold)
cv2.imshow("mask2",mask2)
and1=cv2.bitwise_and(img,img,mask=mask1)
cv2.imshow("and1",and1)

and2=cv2.bitwise_and(and1,and1,mask=mask2)
cv2.imshow("and2",and2)

lowest=np.array([73, 145, 45], dtype=np.uint8)
highest=np.array([114, 255, 255], dtype=np.uint8)

mask3=cv2.inRange(and2, lowest , highest)
cv2.imshow("and3",mask3)
and3=cv2.bitwise_and(and2,and2,mask=mask3)
cv2.imshow("and3",and3)
gray=cv2.cvtColor(and3,cv2.COLOR_BGR2GRAY)
cv2.imshow("gray",gray)

gaussian=cv2.GaussianBlur(gray, (3, 3), 5)
cv2.imshow("gaussian",gaussian)

edge=cv2.Canny(gaussian, 10, 150)
cv2.imshow("edge",edge)

rho = 1

# 1 degree
theta = (np.pi/180) * 1
threshold = 15
min_line_length = 20
max_line_gap = 10

lines=cv2.HoughLinesP(edge, rho, theta, threshold, np.array([]),
                     minLineLength=min_line_length, maxLineGap=max_line_gap)
for line in lines:
    for x1,y1,x2,y2 in line:
        cv2.line(img, (x1, y1), (x2, y2), [0,255,0], 1)
        m=(y2-y1)/(x2-x1)
        print ("y="+str(round(m,2))+ "x+"+str(y2-y1))

cv2.imshow('output',img)

cv2.waitKey(0)
cv2.destroyAllWindows()
#cv2.imshow('output',img)
cv2.waitKey(0)
cv2.destroyAllWindows()
Manas Sikri
  • 23
  • 4
  • Can you post an example of a failure case? It's hard to advise you without knowing how variable the input images are and what sort of failures you are experiencing. – Mozglubov Nov 15 '18 at 21:19
  • @Mozglubov I have added one failure case.In that case the lines or the boundaries are not being formed properly over the white penalty boundary lines. Thank you – Manas Sikri Nov 16 '18 at 17:45
  • Ah, okay, so the images are always from roughly that camera angle, and it's more an issue of the thresholds? – Mozglubov Nov 16 '18 at 18:34

0 Answers0