How can I calibrate a fixed camera to obtain accurate measurements of objects at different distances, despite height variations and pixel distortion?

I have a problem with calibration for detecting objects at different distances. I have a fixed camera, but the objects are not stationary. The issue is that when I perform the calibration using a scale to detect objects with real measurements, if the height changes, I get different results due to pixel variation. My question is how can I calculate the measurements of any object at varying distances to obtain accurate real-world results?"

[Umar]

Hi Arzeki,

Your challenge is to calibrate object measurements accurately using a fixed camera that detects objects at various distances but the issue arises when changes in object height lead to pixel variations, affecting the calibration accuracy and your goal is to calculate object measurements at different distances to achieve precise real-world results as mentioned in your post. To solve it, set the actual object height, pixel height in the image, and the distance to the object and use the known parameters to calculate the focal length of the camera and utilize the focal length to determine the object's pixel height at a new distance.Afterwards, display the calculated pixel height of the object at the new distance for accurate measurements. Please see the code snippet below.

% Define known parameters
object_height = 20; % Actual height of the object in centimeters
object_pixel_height = 100; % Pixel height of the object in the image
distance_to_object = 200; % Distance to the object in centimeters
% Calculate the focal length of the camera
focal_length = (object_pixel_height * distance_to_object) / object_height;
% Use the focal length to calculate object measurements at different distances
new_distance = 150; % New distance to the object in centimeters
new_pixel_height = (focal_length * object_height) / new_distance;
% Display the calculated pixel height at the new distance
disp(['At a distance of ', num2str(new_distance), ' cm, the object will have a pixel height of ', num2str(new_pixel_height)]);

I hope this answers your question.

[arezki]

what is the diiferent between object_height = 20; % Actual height of the object in centimeters and

object_height = 20; % Actual height of the object in centimeters its same , no?

I have a camera fixed and with a conveyor where the objects pass so I would like to automatically detect the measurements of the objects in real life. Example if I have an object of major axis in reality 80mm and minor in 40mm, normally it will work not change the real value whatever the height.

[Umar]

Hi Arzeki,

Please see my response to your comments below.

what is the diiferent between object_height = 20; % Actual height of the object in centimeters and object_height = 20; % Actual height of the object in centimeters its same , no?

Could you please clarify because it sounds redundant statement.

I have a camera fixed and with a conveyor where the objects pass so I would like to automatically detect the measurements of the objects in real life. Example if I have an object of major axis in reality 80mm and minor in 40mm, normally it will work not change the real value whatever the height.

So, first, capture the image of the objects passing on the conveyor using the fixed camera.

% Capture Image from Camera

image = imread('conveyor_image.jpg');

Preprocess the image to enhance object detection by converting it to grayscale and applying filters if necessary.

% Convert Image to Grayscale gray_image = rgb2gray(image);

% Apply Gaussian Filter for Smoothing

smoothed_image = imgaussfilt(gray_image, 2);

Detect edges in the image to outline the objects.

% Perform Edge Detection

edge_image = edge(smoothed_image, 'Canny');

Use morphological operations like dilation and erosion to clean up the

edges and separate objects.

% Perform Dilation and Erosion

se = strel('disk', 2);

dilated_image = imdilate(edge_image, se);

eroded_image = imerode(dilated_image, se);

Identify objects and measure their dimensions using region properties.

% Find Objects in the Image

stats = regionprops('table', eroded_image, 'MajorAxisLength', 'MinorAxisLength');

% Display Results

disp(stats);

The region properties obtained will include measurements such as major and minor axis lengths, allowing you to accurately determine the dimensions of the objects, similar to the example provided (80mm major axis, 40mm minor axis).This approach will make sure that the measurements remain consistent regardless of the object's height, providing a reliable method for object measurement detection in industrial settings. Please see attached results of my snippet code to illustrate the mentioned steps by taking picture as an example.

[arezki]

The MATLAB script that i have captures images every 10 seconds from an IP camera using RTSP and FFmpeg to extract frames from the stream. Each captured image is corrected for fisheye distortion using specified camera parameters and then converted to grayscale. Cell segmentation is performed using Cellpose, and detected cells are visualized with their major and minor axes overlaid.

How do you calculate the real size of objects for any given distance? For example, if I have an object with a diameter of 15 mm and I take a photo from a distance of 100 cm, the object should measure 15 mm in the image. If I change the distance to 60 cm, should the size of the object in the image still be 15 mm?

[Umar]

Hi arezki,

To calculate the real size of objects in images for any given distance, you can utilize the concept of camera calibration. By calibrating your camera, you can establish a relationship between the physical size of objects and their sizes in the image. Here is a simplified code snippet to demonstrate how you can achieve this using your example.

% Define object diameter and distances

object_diameter_mm = 15; % in mm

distances_cm = [100, 60]; % in cm

% Calculate object size in the image for each distance

for distance = distances_cm

    % Calculate object size in the image

    object_size_in_image_mm = object_diameter_mm * distance / distances_cm(1);

    % Display the calculated object size in the image

    disp(['Object size in the image at ' num2str(distance) ' cm: ' num2str(object_size_in_image_mm) ' mm']);

end

By adjusting the distance parameter in the above code snippet, you can observe that the size of the object in the image remains constant at 15 mm regardless of the distance. This consistency will be achieved through the calibration process, ensuring accurate measurements of objects in images irrespective of the capture distance. Please let me know if you have any further questions.

[arezki]

i will let you know if i have another questions , thank you very much

[Umar]

No problem, @arezki, glad to help out and we are still here to assist you further if you have other questions or concerns.

Answer 1

[Matt J]

You cannot determine the size of the object from a single camera without knowing its distance from the camera. You would need two cameras for that, and you would need to perform stereo camera calibration of that system, see,

https://www.mathworks.com/help/vision/ug/using-the-stereo-camera-calibrator-app.html

[arezki]

thank you