Object Tracking from scratch with OpenCV and Python 

Can we do this complete process in raspberry pi

I need the sources code

@@Abbababagana-sw6wd can share me source code?

Hi, I applied yesterday with my other to get contents, but still no response. Can you pls tell me how long it takes?

Yeah, he is the best

The main component to run smoothly is an Nvidia Graphic card. It should be Nvidia GTX 1650 minimum, then a standard CPU Ryzen 5 or Intel I5 and minimum 8GB of RAM. Also the it's important that the Opencv library should be installed with GPU.

exactly, this only works with Opencv 4.4.0 or greater, so I guess you're using an older version. Second thing to know, raspberry pi is too weak to handle object detection in realtime with YOLOv4, so you should integrate it with a smaller algorithm.

@@pysource-com thanks for the advice. maybe I'm hoping for too much out of the RPI.. I'm trying to make something that lets me know if someone is coming in to my gate or going out of it. one or two FPS would do probably and the video isn't being viewed or used for anything else as I have other HD cams for that .. so resolution etc isn't important.. Just an LED for going in or going out.. When I followed your tutorial on my PC it works fine.. my desk top is a little big to stick on my gate post though :-) Thanks again..

@@michaelallen185 yeh did u run it on GPU need help

@@michaelallen185 thanks buddy for help

that's normal to be that way. If you have a CUDA gpu you can install Opencv CUDA to use your graphic card and it will be much faster. on windows though, it's a pretty complex installation

I'm not 100% sure on this, but since YOLO produces output based on class, I'm sure there is a way you could classify bounding boxes by class, and then track in that way.

​@@noahvandal6485 i have a idea now, i can performe a motion detection excludind de bb of yolo, or maybe edge detections with dimensions. My question was more about of how to detect objects that is not part of YOLO classes.

It's normal, if Opencv is not installed with GPU, it can process around 1 or 2 frames per second only. You would need an Nvidia GPU and you would need to configure Opencv in order to work with that.

@@pysource-com Thank you for you answer

Hi, can you respond me please😌

hi, this is only available on the videocourse Object Detection with opencv and deep learning that you can find on pysource.com , the section courses.

yes, it is a custom file I created. you should download that from the blog post related to this video. Links are in the description

hi, I encountered the same error and downloaded the file as you said. But it still gives the same error. Do I need to change and rewrite the code? Did you solve this problem?

I guess you solved this problem already because you asked that one year ago but maybe someone watching that video has the same question. You can import streams via: cv.VideoCapture(0) with incrementing numbers to iterate over hardware inputs If its a stream via ethernet you would have to use a thirdparty library to import the stream (thats more complicated but can also be done).

hello. i am working on these subject for 5 years and i can say unfortunately there is not a way to run these things on raspberry pi like devices with high fps. i tried even jetson nano but if you use deep learning based approach to detect objects you have no chance. i am using dlib and opencv to detect and track faces. dlib is very good choise to detect faces accurately. and opencv's mosse tracker is really really fast. so with cpu overclocking on rasperry pi 4 i am getting 20-25 fps. but remember these devices (raspberry pi, beaglebone, odroid etc.) are not good choices to run deep learning solutions. by the way raspberry pi is the best choice in these devices. asus tinkerboard is the worst.

@@ycdcherokee What if I implement TensorFlow Lite and Edge TPU on my Raspberry Pi? Please note that I have no knowledge of the program. I'm trying to do simultaneously worm brain neurosimulation and object recognition an IoT device like the GoPiGo3. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-9kxty2qec4g.html

That would require mapping the area and it's much more complex than this. You also should have depth cameras or lidar sensors to get coordinates from the real world, you can't do that from a simple video.

@@pysource-com Got it, thanks a lot for your response. Is there any way to normalise these coordinates?

Make sure you hav visual studio installed and add C/C++ extensions in VS

pip install opencv-contrib-python

I have an Nvidia GPU and my Opencv is installad with GPU support. I don't show this on the video, as it's mostly an exercise for beginners to understand about object tracking

@@pysource-com I tried to install Cuda but the nvidia driver installer fails everytime, I unchecked the Visual Studio Integration and Geforce experience but it's still no use. Im using window 11 and my card is geforce mx230. is it because my card is incompatible?

I met the same issue. Have you solved it?

Source code is free to download from video description. There is the blog post related to this video (and inside you'll find the download link)

@@pysource-com .. Thank you for your response. Please what is your charge for consultancy on a project? I have a project I am curently working one which I would like your input in?

Plus, where I can't see your object detection video in your opencv with python playlist. You mentioned in this video that you already did an object detection video? Could you please send me the RU-vid link for the object detection video... Kind regards

@@babatundeoladapo3929 you can contact me here pysource.com/contact/

You can do it either way, it's up to you. but if you're using pycharm i recommend to install it already with pycharm

@@pysource-com yeah I have some libraries installed with phycharm.

The application I use to write the python code is Pycharm

you need to make sure that Python is installed and executable from the windows CMD. Instead of pip you can try running this: python -m pip install opencv-contrib-python

It should be working correctly now. Contact me here pysource.com/contact/ if still you're not able to download the files

I'm not familiar with any existing codes that do that, but it's something you can certainly program from scratch with Opencv.

@@pysource-com Thank you another question please In one of the videos, I took one area of ​​interest to count the cars if I wanted to do the same thing, but on two areas of the video so that the cars were counted together in the two areas, and the car numbers in the output were distributed over the two areas as you called them "roi"

@@pysource-com ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-O3b8lVF93jU.html

That is an interesting project that can help in the agricultural field, so I might do that in the future

That's not an error, that's normal with the standard installation of Opencv. Simply if you have a graphic card Nvidia GTX with 4GB+ of RAM, you can compile Opencv to use the GPU and get more than 10 times faster speed. But this is not for beginners, it's a more advanced steps which is not important for this tutorial.

@@pysource-com Thank You for the quick reply Sir. But I'm getting 1 fps. Is there any workaround for this?

​@@Faisalamin01 got the same issue

@@yidirbg93 read his reply, it's just a warning not an issue

@@Faisalamin01 but my issue is that i want to analize a live video, but i can't because the object detection takes too many time if I wanted to analyze a pre-recorded video, there would be no issue, but that's not what i want

Same problem bro...

now what should we do??

That's normal, it's not an error, more of a warning. Opencv runs by default with CPU. If you have an Nvidia GPU you can compile it with that to speed up the detection.

But video is playing in 1FPS what should we do???

@@pysource-com yes could you please help how to compile it with GPU?