Really nice project!!! The only thing is that you are still running object detection on every frame, which not only detects new objects, but also previously detected ones. This makes the code impossible for real time applications, unless you are using some device with accelerator. Why don't you make use of object tracking to avoid detection on every single frame? This could speed up the application and make it usable in real time.
Skipping frames won't make the tracking reliable. The more frames you can handle, the better. A computer with a decent Nvidia GPU will handle this in realtime, same is with the Nvidia Jetson Xavier.
Great demonstration.. Is it possible to do a video where you talk about the business side of doing computer vision ? For example, setting up a business to help businesses use computer vision in my local region
That is interesting! I'll probably do that at some point as I see very valuable and with a huge potential helping local businesses with Computer Vision
thanks so much your DeepLearning to help me very much. Now am using the same method or coding you using for vessel to detect peer,before berthing alongside to avoid collision and damaging the outside of peer and vessel as well! how to applying this coding? please need your help.... many thanks indeed!!
in this case you need to take into account more parameters: - get the current vessel's speed - calibrate camera view to get pixels distance in meters distance Tracking the vessel, knowing the peer position, speed and distance you'll have the information necessary for your project.
@17:19 Jetson Xavier and jetson nano for Sort Algo but for Deep-sort Algo what kind of Procesessor are required? can we run Deep Sort Algo on Raspberry Pi 4?
Hello! In 19:21 , is there a way to make the ROI be that exact shape like in 20:15 ? I want to be able to fit the ROI perfectly to the road, but using frame[x1: y1, x2, y2] always gives me an ROI that is a rectangle.
Thank you, sir, that's really amazing! I have a question, what if i have two different cameras for the same project of object detection and tracking, what well be the best way to do a comparison between counted objects whether camera 1 has more objects or camera 2?
If the cameras have a common area, would be a good option to link them through panoramic view, otherwise it would require a more sophisticated comparison method, for like for example facial recognition, or plate reading for cars.
Why does yolo detect exactly same object on exactly same frame but with different probability, from 11:52? I mean even if the video has low fps, and yolo makes detections in between, these detection are made on exactly the same frames (images). Shouldn't scores from the detection model be always exactly the same on exactly the same images?
Good observation. This video is slightly abnormal, as I recorded it by recording the screen at 30 FPS when the video is has lower FPS than that, so there are a few frames almost identical to them. The reason why the score changes is because even if the frames are almost identical, they're not 100% the same, but a a few pixels inside and some shadows are slightly different during the transition. With normal videos with the right amount of FPS, or even better taking the frames from the camera in realtime everything will be more smooth and precise
Changes in light may cause several frames to fail to detect the same object during object detection. There are still many problems that need to be resolved.
Hi, this specific video is only a demostration, not a tutorial, so source code is not provided here. Source codes for tracking are provided only inside the course Object Detection with Opencv and Deep learning at pysource.com
Nope, raspberry pi is too weak for real time image processing with Artificial intelligence. The minimum required is Nvidia Jetson nano, but even better would be the Jetson Xavier. For faster processing and with multiple cameras it's necessary a computer/server with RTX or better Nvidia GPUs
🔥Learn how to build your own AI vision solutions: pysource.com/community
Great job. A good teacher, but the Channel is "Pysource" without the Python source program !?
I was exactly waiting for how to apply YOLO on certain area of the video during the day until i find ur video before i am about to sleep. Thx a lot!
awesome video bro, could this work on a live camera?, for example using an ESP32 Cam? . thanks for the explanation lad!
Really nice project!!! The only thing is that you are still running object detection on every frame, which not only detects new objects, but also previously detected ones. This makes the code impossible for real time applications, unless you are using some device with accelerator.
Why don't you make use of object tracking to avoid detection on every single frame? This could speed up the application and make it usable in real time.
Skipping frames won't make the tracking reliable. The more frames you can handle, the better.
A computer with a decent Nvidia GPU will handle this in realtime, same is with the Nvidia Jetson Xavier.
good job
Hi, I do not have deep_sort_v3 file that you imported in the code. Is it possible you could upload it?
Great demonstration.. Is it possible to do a video where you talk about the business side of doing computer vision ? For example, setting up a business to help businesses use computer vision in my local region
That is interesting! I'll probably do that at some point as I see very valuable and with a huge potential helping local businesses with Computer Vision
Could you please share code of Object Detection class and Deep Sort you are importing?
I try to find deep_sort_v3 to 😊
Thank you for the video it is really good. Could i access the project codes somewhere to look at it clearly?
thanks so much your DeepLearning to help me very much. Now am using the same method or coding you using for vessel to detect peer,before berthing alongside to avoid collision and damaging the outside of peer and vessel as well!
how to applying this coding?
please need your help....
many thanks indeed!!
in this case you need to take into account more parameters:
- get the current vessel's speed
- calibrate camera view to get pixels distance in meters distance
Tracking the vessel, knowing the peer position, speed and distance you'll have the information necessary for your project.
@17:19 Jetson Xavier and jetson nano for Sort Algo but for Deep-sort Algo what kind of Procesessor are required? can we run Deep Sort Algo on Raspberry Pi 4?
Special Thanks for you tutorials 👏👏👏🙏
Hello! In 19:21 , is there a way to make the ROI be that exact shape like in 20:15 ? I want to be able to fit the ROI perfectly to the road, but using frame[x1: y1, x2, y2] always gives me an ROI that is a rectangle.
Thank you, sir, that's really amazing!
I have a question, what if i have two different cameras for the same project of object detection and tracking, what well be the best way to do a comparison between counted objects whether camera 1 has more objects or camera 2?
If the cameras have a common area, would be a good option to link them through panoramic view, otherwise it would require a more sophisticated comparison method, for like for example facial recognition, or plate reading for cars.
Hi,
Please do you have deep_sort_v3 file ? if yes can you please share it with me please?
Thx 😊
Where I have to do coding
Like collab etc...
Please give some information
Could you share a download link for the MP4 file you used on this video?
What if the coordinates I want are not on the picture, but outside the picture
please , where can I find the deeb sort library or file ? from where can I download it ?
do you find it ?
Why does yolo detect exactly same object on exactly same frame but with different probability, from 11:52?
I mean even if the video has low fps, and yolo makes detections in between, these detection are made on exactly the same frames (images).
Shouldn't scores from the detection model be always exactly the same on exactly the same images?
Good observation. This video is slightly abnormal, as I recorded it by recording the screen at 30 FPS when the video is has lower FPS than that, so there are a few frames almost identical to them.
The reason why the score changes is because even if the frames are almost identical, they're not 100% the same, but a a few pixels inside and some shadows are slightly different during the transition.
With normal videos with the right amount of FPS, or even better taking the frames from the camera in realtime everything will be more smooth and precise
@@pysource-com ok, I understand now, also I think that maybe the effect of compression of the "outer" video is changing some pixels slightly.
@@grzegorzkozinski2308 exactly
Sir can you make tutorial for Yolo object identification with voice ? Pliss Sir
Thank you sir, Great Video tutorial
Hi sir, I want to work with this for a project, can u please assist me
Man can you make a video on how to make a 360 cctv camera follow a specific objective. With python
Thanks sir, this a great content
Changes in light may cause several frames to fail to detect the same object during object detection. There are still many problems that need to be resolved.
By training the Deep Learning model on the specific scenario, with different lightening conditions, the detection would work well regardless.
Good offer, but why you steel sergio from all of us.
Let him to teach us Computer Vision.
Hi,can you make a video how to train custom data set in YOLOv4?,thank you
I might do that
@@pysource-com Please do that,nobody explaining it like you.You are explaining it easily.
Thanks
Thanks sir, this a great content
Where i can find code source please
Another amazing tutorial!! Thank you!!
Nice video, thanks.
opencv cannot open url with https !
Wow really Great, I'm getting lots of informations, Thank you so much for the Awesome sessions🥰🥰🥰
can the code work on real time web cam feed??
Where can I find yolo_detection library?
Thank for you amazing tutorial
thanks for this amazing video
Sir i want to follow this program but i can't see from yolo_decetion import * from the file Sir
Hi, this specific video is only a demostration, not a tutorial, so source code is not provided here.
Source codes for tracking are provided only inside the course Object Detection with Opencv and Deep learning at pysource.com
@@pysource-com OK, how much?
Hello sir,
Thanks for this great video.
Please can you share the deep_sort_v3 file with us please?
Thx ,merci, gracias, chukran, thanmirth 😊
Have you found it yet?
ya lo encontraste ???
THX 👏👏
inference is very slow.
Cant impelemntation in raspberry Pi ?
Nope, raspberry pi is too weak for real time image processing with Artificial intelligence.
The minimum required is Nvidia Jetson nano, but even better would be the Jetson Xavier.
For faster processing and with multiple cameras it's necessary a computer/server with RTX or better Nvidia GPUs
@@pysource-com which get turorial them ?
great video, please could you share de code of deepsort_v3 :)
Hi , please i am working on it i have a project i am a student can you help me please with a program python ?
Can i have your email