Ziri's profile - activity

Take a look at the generalize hough for edge based template matching . you can also try SIFT

Take a look to the generalize hough for edge based template matching . you can also try SIFT

This is not an opencv problem ... you'll need some edge based template matching (rotation invariant). If your camera is

This is not an opencv problem ... you'll need some edge based template matching (rotation invarinate). If your camera is

Did you run any "Unet" program ? The output should look something like that.

Search for annotation tools online - > the goal is to have 2 images 1 is the original and the second is the a binary

If you use Unet you' ll be able to get the tress as 1 segment ; you'll be able to get their borders& position(centr

If you use Unet you' ll be able to get the tress as 1 segment ; you'll be able to get their border& position(centro

I hope you have high resolution images like the second one. To make it simple i would run an edge detection/contour de

You can threshold to separate them then create you own OCR version by training those images. (Or retrain tesseract).

If you've successfully ran that code check blob.h and track where void calculate_blob_center() or get_rect() are ca

You'll need set it as a global variable in the top level of your application : Check documentation here: setUseOptimiz

This is a segmentation task and it seems like deep learning "Unet" gives good results in this case. you can check an ex

This is a segmentation task and it seems like deep learning "Unet" gives good results in this case. you can check an ex

You'll need set it as a global variable in the top level of your application link text

Do you have images database you want to detect trees in ?

You already have the car detection routine ? has bounding boxes ? you can copy the bounding box ROI and save it .

You just need Calibration ( Detect some key points (corners for example) and transform from pixel to world ) + some

You just need Calibration ( Detect some key points (corners for example) and transform from pixel to world ) + some

did you try opening the video without setting any parameters? You can use Gstreamer in Windows Installing on Windows

You have to post your original image before binarizing .

Try Contrast Limited Adaptive Histogram Equalization CLAHE before thresholding.

Try Contrast Limited Adaptive Histogram Equalization CLAHE before threshing.

I am not sure why are you setting these parameters after opening the video : capture.set(cv::CAP_PROP_FOURCC, cv::Video

You don't have to use : cannyImg = cv2.Canny(dialeteImg, 100,200) Just find contours in the binary image then filter

I Suggest this method : 1 ) - as @StevenPuttemns suggested, use watershed or another method to get individual contour

IF you want to check the percentage of the color red(R=255 , G = 0 , B = 0) the count pixels and calculate the ratio.. b

To achieve that you'll need to draw on transparent overlay image instead.

That makes sense now.

Where did you find about: "considering only the 4 corners of an ROI to calculate the mean?"

Accumulate those segments using 2 constraints ( parallelism & distance) OR Collinearity you'll get complete segments

Accumulate those segments using 2 constraints ( parallelism & distance) you'll get complete segments. Then use simpl

I think you're not using Background subtraction the right way. And if you want to use the fixed threshold you'll have t

Thanks . I didn't know that.

Use findHomography to get the right transformation matrix.

For testing... don't train from scratch. Fine-tune an existing model.

You can create large data by changing images you have ( contrast chnage , rotation , translation ...etc)

You can try converting your image into different color spaces and threshold using 3 channels... But if you want more p

The simplest way to do it is to Color threshold (for example) -> extract the blobs -> delete noise -> and find

The simplest way to do it is to Color threshold (for example) -> extract the blobs -> delete noise -> and find

You need 2 point coordinates from your line or segment and your single point coordinates --> then use dot product f

I suggest using DOG or Variance filter to ignore background variation.

Convert your image to HSV or HSL and do simple color thresholding in all 3 channels. If itdoesn't work then segment usi

you'll be able to do the same with Hough but tuning parameters for different cases is difficult.

If you detect contours as a tree with RETR_TREE your triangle becomes a hole. (check contour Hierarchy) you'll also nee

If your purpose is detecting (rectangle with triangle) then doing contour analysis and detecting shapes is better for th

Hi , I added original image (curve shape and orientation are random ) . Thank you.