Double pointer instrument panel reading (I)

The contents of this chapter ： Think about the project and identify the dial     

  I basically put CSDN I've read all the articles and downloads about pointer reading on , Students with needs or questions can come to communicate , Resources can also be given free , The final recognition effect is shown in the picture in the previous article , After optimization, the resources occupied by running the program are particularly small ,GPU not essential . It should be easy to write a paper and get a patent later .

development language :python3.0+

Call package ：

1.sys                  Used to exit the program :sys.exit()
2.time                 Used to calculate time , Also used matplotlib Timing of drawing 
3. win32api            Used to monitor the button   win32api.GetKeyState(27)
4. cv2                 Main call , Most processes require opencv, Version is 3.4
5. imutils             It is mainly used to calculate the coordinates of the center point 
6. matplotlib          drawing 
7. numpy               mathematics 
8. PIL                 Mainly used to transform arry Array as image 

See the following table for the specific version , Too lazy to match one by one , Find it yourself

Talk about the idea of just getting the demand and the process of negation later :

Get the demand : A dial , Identify speed and signal , The dial is round

idea ：hog+svm Training models , Real time tracking of dial position , Hough line recognition , Identify with the corresponding angle

practice ：hog+svm The speed can , But the effect of Hough line detection is particularly poor

reason ： Because the dial uses an oval pointer , The front end has a small pointer , The recognition effect is not ideal ,

How to deal with it ： Use opencv After binarization of the dial by image processing , Use thinning algorithm （Image Thinning）, Similar to the corrosion expansion seen before , After reading it, you can quickly understand , But with the thinning algorithm, the recognition effect is still very poor , That the road is blocked , Change the plan .

Because the intermediate manufacturer changed a dial , The use of hog+svm Look for the dial and it won't work , Change to Hough circle detection (hough cirlce Transform), Transfer is a college question , You can adjust parameters according to your own needs .

circles = cv2.HoughCircles(
        binary, cv2.HOUGH_GRADIENT, 2, 1000, param1=100, param2=100, minRadius=10, maxRadius=5000)

      But there's a problem , The Hough detection circle has always been carried out, and the performance requirements of the computer are particularly high , Time complexity and space complexity , For processing video, the efficiency is less than , Here I came up with a solution , The first recognized circle is used to cut the image , All circles in subsequent videos are identified according to this cutting area , Greatly reduce the performance loss of the equipment . The camera position is fixed later , This scheme is the best scheme compared with many schemes .

Comparison scheme :1.CSDN A blog program ： Template matching

                2. My previous plan :hog+svm

                3. Hough circle detection scheme is applied to each frame of video

                4.CSDN A download scheme : Forget what plan. Anyway, the effect of my side is not as good as the above comparison effect .

Write this chapter here , Work reason , Continue another day