Thanks, boss.

Nice! Hey, are those your CNC tools in your videos? Do you have a CNC shop?

@@ClaytonDarwin I set up a high end MakerSpace for a school in Shanghai the last five years. Couldn't get back because of Covid--so now I only have my own 3D printer, cricut cutter and Maslow CNC router to play with. Need to find another school to set up here in the States.

Very cool. Did you see my recent SCARA plotter ( kzfaq.info/get/bejne/mtCpgMSitb-UpIU.html ). I was trying to come up with something that anybody could build (i.e. teenagers). I would like to develop that more and do some other education-oriented projects at some point. Working on an ESP32 funboard at the moment (in my spare time).

@@ClaytonDarwin Your plotter is nice. I had just started working with the Chinese Homework Machine and was going to have kids try and make one of them. www.scmp.com/yp/discover/entertainment/tech-gaming/article/3060907/chinese-schoolgirl-caught-using-robot-write It is another simple plotter that costs about $100 over there. A clever mechanism for x-y motion.

I've seen those. Haven't had time to make one.

Good idea. I haven't had time lately to return to this project. But this can certainly be a template for a more complex project like what you are describing.

Thanks. Glad you like it. I'm not sure about an architecture video. I'm self-taught for the most part, so I don't have much theory. I just work in parts, getting each function to work correctly, and then try to combine them into a logical, orderly, and efficient whole. I re-write many times, and I often write my own wrappers and libraries if the ones available don't do what I want. Lots of trial and error. Lots of failures. Lots of reading. Lots of Google queries.

I'll have to go and look at it. Can't remember off hand.

All events are handled by either the key_event() or mouse_event() functions. The mouse event function must be registered as a callback, while the key event function can be called at the bottom of the main loop. The event functions basically read event data a set global variables. The main loop, then uses the variables to decide what action to perform (and also what values the use).

I have never used a phone camera as a webcam, but one of my subscribers said it was possible.

You would have to identify 3 points and then do a little trigonometry. OpenCV has quite a few options for things like that.

Yes. Any distance could be treated like a diameter and multiplied by pi.

Two good ideas. 👍 The lens distortion is fixed, so once you have it calibrated, you should be able to adjust based on the stickers. I think.

Direct connect.

@@ClaytonDarwin sir I am new to this, @6:15 and @6:40, which software prompt is that?

Linux command line or python are the two options.

Did you install opencv for python?

Probably you could use opencv to detect edges/lines, or possibly a closed polygonal shape, and then build your dimensions from that. That's the way I would start tackling the problem.

Like the title says, "try it yourself" and then you will know what it can do. Then you can take the code and improve it.

The unit of measure depends on what measure you use during calibration. In the example I used mm.

Any that will work on your system. I've used several.

I have a USB3 camera somebody sent to me. I think it's 1080p, up to 60fps. There is a video about it. I pit an adjustable lens on it, and it was probably set at about 22-25mm. But pretty much anything will work.

Sorry. Haven't messed with windows for a while. Isn't there a devices list?

Any USB camera will work. Higher resolution allows more accuracy. Also, I like CS-mount cameras because they give you a lot of high-quality lens choices.

@@ClaytonDarwin How do you zoom in the image in your video? Is it optical zoom or just digital one?

In the video I use digital zoom. It's a feature of an OpenCV window. Will zoom to the sub-pixel level. If you change your zoom manually, you will have to re-calibrate your camera.

Sorry for the late reply. You could definitely add edge detection or image recognition to this using opencv. The best results will be had by closely controlling the environment and lighting to get a clear image.

You can change it with an OpenCV parameter, or by changing the camera resolution. Or by dragging the frame corner with your mouse.

@@ClaytonDarwin which parameter of OpenCV

Try googling "OpenCV python change window size".

Accuracy depends on the camera pixel count, the lens, and how well you calibrate.

Sure. Same math, just using different variables.

Whatever you use during the calibration, that's what it is.

I do not know. I only work on Linux.

Hi Mitch. I don't really have any specific references other than the OpenCV docs. I do lots of googling and reading, and then go back to the docs.

Yeah. You're right. I need to get back to making stuff. I've just been really busy with work the last couple of years and not able to spend as much time in the shop. But making goofy robots is what I like to do. Thanks for reminding me.

Unfortunately, I don't have the time to investigate this at the moment.

@@ClaytonDarwin ok, thanks.

No idea. I was using it the other day and it worked fine on my setup. Might be that you have a different key binding for A .

I think you can combine the ideas in this video with the ideas in my triangulation video and do what you are describing. I haven't done it, but I have thought about it.

@@ClaytonDarwin i wanna make it for my final project in school, thanks for your info. But now I'm stuck on how to use a USB camera when I've done put it in port USB, because when I tried your code I can't change the camera set, please let me know which code that I have to change for setting the camera to camera USB, not camera laptop webcam?

You have to edit the code and set the camera number according to your setup. This is an OpenCV setting.

Higher resolution cameras, better camera calibration, better lighting (and contrast adjustments). Those are main factors that affect measurement.

@@ClaytonDarwin I've seen some document cameras on Amazon that apparently compensate for the distortion issue. These have much higher resolution than most web cams and can focus very closely. On the down side, the depth of field is not great, i.e. these are mostly suited for relatively flat objects such as books or other documents.

Interesting. Might work in this situation.

A small reference object assumes that the frame has already been normalised, which is what we are doing with the measurements. Once the frame is normalised, you could use a reference object. Alternatively, you could use a full-frame reference object to normalize the frame. I don't like the one in OpenCV, and I didn't have time to write one myself the day I made this video. This code is just an example, not a product.

Distance from camera generally requires 2 cameras unless the target object is of a know size. Like my triangulation video. Also, this script is real time. You can make it an iterator and yield the data rather than plot it if you need to.

Excellent. Yes. Use it however you like.

@@ClaytonDarwin thanks with full respect dear

The camera is defined in OpenCV by a number. You can change that near the top of the script. It is marked with a comment.

This uses geometry/trigonometry to calculate distances based on a known distance (the distance from camera to table). You can use a similar algorithm to do the measurements you describe as long as you have a known distance to target measurement.

@@ClaytonDarwin You can also build it as relational measurement between height x width / arms x legs. This should be relational to what you are looking for as long as people have all their limbs :) :)

Do you know how to program? You would have to change the code.

When you do the configuration the first time, the program should create a configuration file. This is loaded on start, so you should not need to reconfigure unless the camera setup changes.

@@ClaytonDarwin could you explain more please? what is the type of this file (filename.what?)

@@ClaytonDarwin when I did the configuration, only one excel file that was created, is this what you mean?

Please read the file: gitlab.com/duder1966/youtube-projects/-/blob/master/OpenCV/camruler/camruler.py#L118

@@ClaytonDarwin I found the file, thanks.

There isn't a simple solution for something like that. You would need something similar to the OpenCV library to do the object detection and image manipulation, but a very reduced and streamlined version to run on the ESP32. I haven't seen anything like that. Memory is going to be a problem. Most image libraries want to hold a full image in memory in order to work on it.

Thank you for the quick reply. Then I will try it with the PI 4. Maybe you can make a video how to install everything exactly in what order. Please continue to do so that helps many users who are not so fit with the software

I've never tried it, but I think a Pi4 should work, although you might not get as high of a frame rate as you could on a larger machine.

When you do the calibration the file is created.

@@ClaytonDarwin However, if I use your program on Git Hub, the .cv file already exists. This means that if I rewrite all your programs, so that mean a new .csv file will be created containing pixel and scale data for calibration?

Whenever you do the calibration a new calibration file is created.

The other csv file is the setup variables.

The calibration do you mean is a measurement to get the cal_range value in config mode?

This is just a demo script. You will need to write your own to adapt to a specific task.

I think so. With a single camera you will need a fixed reference point some exact distance from the camera, and then I think you can work out the distances as done in this video using some trig/geometry. You could then convert it to GPS coordinates from a known point.

Please share your code with us.

I don't know the answer. I've never used a phone as a webcam. It will have to be installed in a way that OpenCV can access it as a camera.

@@ClaytonDarwin yeah, i understood due to the way how the code gets the image, i already have something working but not at 100% i would need some extra help on that

Okay. Good that you have it partially working. I'm not able to investigate it at the moment.

@@ClaytonDarwin do you think with would be reliable to aply this kind of system to a company ? Controlling measurements with it

Reliability is based on calibration and testing. Lots of testing. But I would certainly have no issues with using this setup as a base for an industrial application.

The calibration procedure shown remove the distortion.

Ah thank you! 😂 i not aware

@@ClaytonDarwin did you use a telecentric lens for this project?

@@haikalhakim I don't think so.

You should be able to run 2 instances. You'll have to explicitly identify the cameras.

You will have to add object recognition to prevent false alarms, and then insert the alarm routine after the measurement is calculated.

The easiest thing would be to write data to a .csv file. Comma separated values can be read by any spreadsheet software.

Please see the code.

In the main loop, about the same place where it writes the values to the frame, you could also write the values into a database or append to CSV file (imports into excel). There are other options, and you may also want to have something you click, otherwise you will have all the stray measurements going into your data. This was really just a template that you could use to make another application. If you want to automate, you will have to do a lot of filtering to get just the data you want.

@@ClaytonDarwin Thank you so much for the reply, I'll be trying to measure inhibition of bacteria in a culture medium representing a no growth halo, kudos to you, so this video and your project helps me a lot, I can't thank you enough sir, God bless you, peace and love from Iran

One very nice thing about OpenCV is that it has a relatively easy way to train for object recognition. So, you always have the option of adding image recognition into the application. Or maybe more simply shape recognition. This might help you pick your targets automatically. It may be more programming that you are able to do if you are a microbiologist, but if you are in a university or research environment, there is probably a programmer that could help you.

@@ClaytonDarwin indeed sir I am grateful for your advice, I am new to the AI field and I appreciate your work and you sharing your knowledge, God bless you and your family🙏

@Lonesome Herd how did your writing to CSV go, can you share? Thanks

I can't answer that. The software is not making the camera out of focus.

If a 2D area has a direct relationship to weight for the specific objects that you are measuring, then yes. However, this is not the case for random objects.

Latest version of Lubuntu, but any Debian derivative works. I'm using the default Ubuntu install of OpenCV from the time the vid was made.

USB camera, desktop PC, ruler.

Yeah, for 3D you will want 2 cameras, and some type of object recognition. You could use two of this setup, or or one of something like the triangulation video. From either, you can get the dimensions you need.

@@ClaytonDarwin so knowing the distance from the camera will still not provide an accurate measurement? i will need 2 cameras to achieve this?

If you know the distance to the table and the distance to the top of the package, then you have the height. If you compensate the 2D area to be accurate at the top of the package, then you have area. Area times height is volume for regular shapes. It would take some adjustments to be able to get the top area for different heights. So, I guess you could do it with one camera and a range finder, but it may be easier with 2 cameras.

👍 OpenCV has a distance mapping method included.

Sure, but I'm not sure how you would handle calibration and video.

I am also getting this error, im not very adept in python so was wondering if you knew what it was or how to fix it? CAMERA: [0, 1280, 1024, 1310720, 30] key: [97, 'a'] Traceback (most recent call last): File "camruler.py", line 504, in x1c,y1c = conv(x1-(cx),y1-(cy)) File "camruler.py", line 135, in conv scale = cal[baseround(d,pixel_base)] KeyError: 820

I think a raspberry pi will have a hard time with the real time frame rate along with the image manipulation required for measuring and display. However, I haven't tried it on a pi, so I'm not really sure.

I'll have a look at the code in the morning. But basically the dictionary doesn't have the item being indexed.

Sure. You can modify the script to not loop, but instead modify one frame and stop. Just use the OpenCV-Python function to open a file instead of starting the camera. But you will have to go through the same calibration process for the camera you use. Everything depends on the camera setup and calibration.

@@ClaytonDarwin cool, thanks, oh also can it somehow output the detected width and height to terminal?

Yes, just the same as I print other values.

Not really. I did this video because so many people were asking about measuring after watching some of my other videos. I haven't done anything with it since. There are many unexplored options.

@@ClaytonDarwin Thank you for your reply. I will try it myself and if successful, would share the results with you. :)

👍

You have to change the code. Instead of opening a camera you open a file.

@@ClaytonDarwin

@@ClaytonDarwin yeah could you help me with that I'm sort of new to opencv again thanks :D

Most OpenCV Python tutorials start with reading a video file. I'm not sure I'll have time to look at this anytime soon.

@@ClaytonDarwin yeah I figured as well thanks though

OpenCV has built-in functions for color detection. There are many online tutorials for this.

Add some code to save it.

OpenCV has several object detection methods you could add into the mail loop.

Just write the value to file somewhere in the display loop.

Any USB camera should work.

@@ClaytonDarwin i liked this pixel quality. That's why iam asking sir.

I'll check and see what it was.

This one: kzfaq.info/get/bejne/p8xdg66iq87bfZs.html

The script frame_capture.py is a general module that I use for many different application. It works in a separate thread and captures camera frames as they happen. The frames are put into a buffer (holding area) and stored until the main loop can use them. The modules allows the camera to be managed outside the main loop, and prevent frames that are generated by the camera while the main loop is working from being overwritten/lost. In most cases, cameras only have a one-frame buffer. They read a frame and replace the one in the internal buffer, so it you don't read it in time, you lose it.

@@ClaytonDarwin thank you

Hi Tumininu. You can definitely take my script and adapt it. OpenCV can easily handle movement detection and object recognition, and it is straightforward to write the data to a csv file with Python. However there may be some physical problems to overcome. The script I wrote uses a single camera, and this requires that you know the distance from the target to the camera in order to determine the measurement. If you don't know the distance to the target, you will have to have a second camera and use triangulation (like my other video) to get the distance. You can then estimate size. Another problem (maybe) will be the raspi 4. It may not be able to process images fast enough to give you the frame rate you desire. The older raspis were very slow, but I haven't tried it with a new raspi 4. Maybe they are fast enough now.

Thank you Clayton. I installed the OpenCv on my Raaspberry pi 4. It seems its to workon sbb_cv virtual memory. I copied your script from your site into my /home/pi/youtube-projects/OpenCV/camruler folder. i tried running 'python3 camruler.py 0'. it gave error on Line 12 in : No module named 'cv2'. kindly make suggestion. My setup uses 2 USB webcamera with known distance from the object (animal). I forgot to tell you I am a novice with programming. I just have a vision of this project and I am learning on the job.

@@tumininuadebambo raspberrypi-guide.github.io/programming/install-opencv

@@ClaytonDarwin Thanks i ll go through this.

Hi Clayton. Thank for your usual asistance. I got it working. The frame is so large. How can i minimise it. The video seems i am only working at a corner of my whole view. I have tried changing several figures: width, height.... on the camruler.py. Kindly help again. ps dont get tired. cheers Can you also help withh the line to extract the measurement to a CSV file. Thanks

With a scale? Unless you know some way to estimate weight from 2D measurements.

@@ClaytonDarwin Let's say we know the weights of some objects. How do we write it into the program? Do you know what the code is?

Add a look-up table in the same place the area is calculated and added to the frame.

@@ClaytonDarwin Thanks. Could you please write the code?

I generally don't do other people's work without competition.

Nope. But it should translate easily.

Probably don't have time to do that.

@@ClaytonDarwin ok

I haven't been able to get your message translated yet.

I think so, but I don't have time for it right now. Combine the triangulation video with this one.

I am working on it currently. Do you have a solution already?

Because you don't know the distance to the target, 3D measurements typically requires 2 cameras and triangulation.

With this setup you only have two dimensions, so area only.

@@ClaytonDarwin Thanks for your reply. Does using a stereo camera help calculating volume?

Yes.

@@ClaytonDarwin Any suggestions which one works the best, in case you ever tried it.

I have a video where I do some work with two cameras and triangulation to determine distance. However, I haven't tried any off-the-shelf 3D software or cameras.

And what kind of microcontroller did you used? Thank you, Sir, in advanced.

This video does not use a microcontroller.

To run this code you will have to install Python and Python-OpenCV. Then it should run using a standard web camera over USB.

@@ClaytonDarwin I already have it, Sir. Sir, can you explain your codes that you used in this video for the beginner like me? I mean, do you have tutorials video on how you write those codes?

@@ClaytonDarwin If I will make a prototype on auto-measuring, what microcontroller you can suggest?