Video capture/processing library recommendations

[David P 3]

Hello

 

I would like to be able to connect to a video camera (USB Initially) to capture footage and to process the footage (frame by frame), either live or after it has been saved.  The idea is to create a traffic counter.

 

Can anyone recommend a video capture/processing library

 

Many thanks

[Cinemaizer 3]

See https://www.trichview.com/features/rvmedia.html

[Tom F 85]

We have used https://www.mitov.com/products/videolab#overview for 13 years.  Excellent library to which I owe my company's product's success on Windows and macOS.  His tools are threaded, which means you get maximum through-put and the VCL doesn't freeze.

Free trial and free to non-commercial organizations. 

The installer puts some demo programs here C:\Program Files (x86)\Embarcadero\Studio\22.0\LabPacks\Demos.  You'll find one there that lets you process frame by frame either a live feed or file playback.  Just drop a few components on a designer, hook them up, and grab the "Frame Ready" event to process your data.

Check out some of his other tools. He has some real-time tools that do pattern recognition and may be all you need.

Boian Mitov is brilliant, IMO.  His tools are the best we could find. No royalties necessary.  No confusion about FFMPEG patent licensing, which we always worried about.

 

The attached sample has a video feed (of a telescope spectrometer) on the left and some data extracted on the Mitov plot on the right.

Tom 

 

 

Edited September 23, 2022 by Tom F
Replaced static image with gig

[David P 3]

Thanks Tom. that's very useful.  Always good to get a recommendation from someone who uses the library.

[Rollo62 539]

On 9/19/2022 at 10:36 AM, David P said:

to create a traffic counter.

Video processing is once thing, but maybe you are looking for object detection algorithms like Yolo too.

https://www.delphipraxis.net/1428590-post11.html

I'm not sure if VideoLab implements that or would let you support that, that would be good to know.

 

One thing to consider is maybe Python with OpenCV too, the Python environment could make such task simple, but I'm unsure if that has enough performance for realtime analysis

https://medium.com/@MrBam44/yolo-object-detection-using-opencv-with-python-b6386c3d6fc1

https://towardsdatascience.com/yolo-object-detection-with-opencv-and-python-21e50ac599e9

https://opencv-tutorial.readthedocs.io/en/latest/yolo/yolo.html

https://pyimagesearch.com/2018/11/12/yolo-object-detection-with-opencv/

[Tom F 85]

Quite a few years ago I played with Mitov's Vision Lab, which is part of his suite.  I never needed to use it, but it was pretty exciting for me to see what it offered for object tracking. There are some downloadable demos here: https://www.mitov.com/products/visionlab#screenshots 

Edited September 23, 2022 by Tom F

[Rollo62 539]

I have tested Mitovs libraries too and they offer great stuff for video processing.

But a complete traffic counter would be not easy to do, last time I checked the libraries only included some standard filters, like Hough, Canny, simple MotionDetection, etc.

This can be maybe helpful to build a simple traffic counter, but I would assume that this performance will be not too good.

Thats why something like Yolo has been invented, also because this is much faster and stable.

 

I have tried some simple object detection algorithms of more or less static images some years ago and I can tell you that the normal image filters cannot magically give good results.

It is very tricky to make them return stable and error-free results, especially if the scene is highly dynamic like traffic and if it can have different lighting situations.

If you have some good ideas or solutions, would be great to see that too.

Edited September 23, 2022 by Rollo62

[David P 3]

Thanks Rollo/Tom.  This is intended as a hobby project intended for use in my local area.  I'm hoping this means I can simplify the configuration/detection a lot.

[Tom F 85]

Hi, David,

ImageEn may be more than you want to spend on a hobby project. (It was a few hundred dollars several years ago when I bought it.)

But the announcement below is pretty exciting. Whether you use it or not, I wanted to post this somewhere here so that others who might be interested would see it too, 
 

Quote

 

We have just released ImageEn and IEVision, our advanced tools add-in.

With ImageEn 11.4.0 you can now use the  Google Vision API to recognize objects, text, handwriting, faces, famous landmarks and logos in images.