5

I am creating system for reading value from 7 segment display pressure meter. For detecting digits and values from digital( Seven Segment Display) meter, i gone through following links:

  1. https://hackernoon.com/building-a-gas-pump-scanner-with-opencv-python-ios-116fe6c9ae8b
  2. https://www.pyimagesearch.com/2017/02/13/recognizing-digits-with-opencv-and-python/
  3. https://www.kurokesu.com/main/2017/02/20/dumb-thermometer-gets-digital-output/
  4. https://www.calloftechies.com/2017/12/character-recognition-explained.html

As per explanation from above links,I have created following steps for reading Digits:

  1. Convert Image to Gray Scale Image.
  2. Generating Binary images using Otsu's Algo for Threshold and Binary Inverse.
  3. Applying Canny Edge Detection using Otsu's threshold value.
  4. Find FindContours on Canny Edge Detection.

This is working good with some images, but when dial has some light color then this is not working.

Original Image with light color. Original Image Gray Scaled Image enter image description here Threshold Applied Image with Otsu's and Binary Inverse enter image description here

You can see that there is no big color difference between Panel display section and Gauge Area so Otsu converts entire dial to whitish ignoring Panel Section

Please also see below images: original, GrayScale and Thresholded. enter image description here

enter image description here

enter image description here.

Please help me for figure out rectangle Thanks in Advance.

Vishal Patel
  • 1,715
  • 16
  • 26
  • Have you considered not using otsus and instead trying find a fixed threshold that works for all of your cases? – T A Nov 25 '19 at 13:22
  • 1
    @TA: "a fixed threshold that works for all of your cases": this never works. –  Nov 25 '19 at 17:17
  • @TA : I had tried with Fixed threshold, but as all images have not same light and angle, it is not worked for all images. – Vishal Patel Nov 26 '19 at 05:08

3 Answers3

3

Here's an approach using contour approximation + contour filtering:

  • Convert image to grayscale and perform Canny edge detection
  • Perform morphological close
  • Find contours and filter using contour approximation
  • Extract ROI

Here's the detected display

enter image description here

Since we already have the bounding box coordinates, we extract the ROI using Numpy slicing

enter image description here

import cv2

image = cv2.imread('1.jpg')
original = image.copy()
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
canny = cv2.Canny(gray, 20, 200, 1)

kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
close = cv2.morphologyEx(canny, cv2.MORPH_CLOSE, kernel, iterations=1)
cnts = cv2.findContours(close, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
for c in cnts:
    peri = cv2.arcLength(c, True)
    approx = cv2.approxPolyDP(c, 0.01 * peri, True)
    area = cv2.contourArea(approx)
    if len(approx) == 4 and area > 1000:
        x,y,w,h = cv2.boundingRect(approx)
        cv2.rectangle(image, (x, y), (x + w, y + h), (36,255,12), 3)
        ROI = original[y:y+h, x:x+w]

cv2.imshow('image', image)
cv2.imshow('ROI', ROI)
cv2.waitKey()
nathancy
  • 42,661
  • 14
  • 115
  • 137
  • Thanks for reply. I got the result, but it works only for this image, you had applied, for other image same algo doesnt work. – Vishal Patel Nov 26 '19 at 07:34
1

Otsu is wonderful at finding a good threshold where there are two dominants intensities. But when there are three of them (white, gray, black), it cannot guess what you want to separate.

If you apply Otsu in a window that rejects most of the background, it works fine again.

enter image description here

0

You may use circle detect first and then digit detect. After getting the results of digit detection, you can estimate the area with high precision. You may using yolo method as a detector.

ouflak
  • 2,458
  • 10
  • 44
  • 49
energy1991
  • 1
  • 2