@@sergeynuzhny9523 Thank you It's very impressive. Do you have any technical document or paper or guide to do it in more detail? It could be very helpful to apply to aquatic farming.
Yes, You can use this project on Jetson - the TensorRT backend for YOLO is included. But in real time on Jetson Nano it works only with yolo4-tiny or yolo5n models
@UCrhac1bb8Ze_u-ViIHFDXoA расскажите подробнее как считается скорость? В статье на медиуме, на которую Вы ссылаетесь, я не совсем понял как считается скорость в рамках белой области. И не смог найти именно для этого код в репозитории
Скорость считается не в рамках белой области а на всём кадре. Белая область показывает 4 точки на кадре, которые были руками поставлены в соответствие географическим координатам на карте (наример в google earth). Затем эти 4 пары координат (гео - пиксели) ручками вводятся в исходниках (см. CarsCounting.cpp). А по 4-м парам точек можно найти персективное преобразование из одной лоскости в другую (плоскость поверхности земли в плоскость поверхности этого же участка Земли на кадре). Таким образом, мы можем любую точку на кадре, которая находится на поверхности дороги, перевести в географические координаты. Далее уже дело техники: для каждого объекта есть траектория в пикселях -> траектория в гео координатах -> считаем скорость. Считать скорость можно тоже по-разному, в ростейшем случае по формуле равномерного прмолинейного движения. Но в реальности для получения точных результатов надо будет предпринять намного больше усилий, учесть дисторсию камеры, усложнить модель движения и т.д. и т.п.
Knowledge coordinates are limited. Only the grid-based coordinate system using the geometric transformation using four points is accurate. You can translate the other videos and watch them. I also have a patent. Let's develop it together.
Hi Sergey I want to use this code to record the path and distance traveled by Mice in tests like this vid:ruclips.net/video/GebcshN4OdE/видео.html but in yolov4 since that video's actual software doesn't work as intended. I already have a trained model that works on different kinds of mice and is able to detect it on yolov4 using a collab notebook made by The AI guy. Thanks :)
@@youtubeu4005 the github code in the link has been updated and works properly if you even want to ask the github creator a question he'll answer as well
@@youtubeu4005 Idk about the collab code thing, I can just give you this repo:github.com/colinlaney/animal-tracking it works the way you are referring to but from a computer vision approach, not a machine learning one, there's also this link to a supposedly working software that deals with these kinds of tests (have not tested it yet because I'm working on the code of the first link to make it work on the y-maze test): github.com/healthonrails/annolid other than those two links there's a youtube channel called the ai guy that deals mostly with yolov3/4 ps: if the 2nd link works please tell me how because I'm putting it on hold for now until I finish the y-maze
I have never working with an animal tracking. I think that it need a lot of work: videos, dataset, testing. I dont know how I can help you with this project
absolutely awesome! very useful technique based on object tracking
Nice. Thanks for sharing and the code.
One question. Where did you get the lat/lot coordinates of the clip from ?
Thnx!
I have founded place from original video ("AIRE DE SERVICE DE L'ABIS") on google maps.
Amazing! Well done
Is it real-time speed or just the speed of a vehicle from line 1 to line 2 (i.e., distance(line1, line2)/ time) ? Thank you!
It's realtime. Lines used only for vehicles counting
@@sergeynuzhny9523 Thank you
It's very impressive. Do you have any technical document or paper or guide to do it in more detail? It could be very helpful to apply to aquatic farming.
@@hoangng16 In description to video I give link to the paper with this approach
Hi! That's very awesome. I want to run it on my jetson nano, can it be run with tensorrt? What is the approximate speed?
Yes, You can use this project on Jetson - the TensorRT backend for YOLO is included. But in real time on Jetson Nano it works only with yolo4-tiny or yolo5n models
hello， can i use this on drones?especially do the speed measuremnet while moving？
This task can be solved with same approach but you need to project frame pixels to relief. In this repository projection not implemented
@UCrhac1bb8Ze_u-ViIHFDXoA расскажите подробнее как считается скорость? В статье на медиуме, на которую Вы ссылаетесь, я не совсем понял как считается скорость в рамках белой области.
И не смог найти именно для этого код в репозитории
Скорость считается не в рамках белой области а на всём кадре. Белая область показывает 4 точки на кадре, которые были руками поставлены в соответствие географическим координатам на карте (наример в google earth). Затем эти 4 пары координат (гео - пиксели) ручками вводятся в исходниках (см. CarsCounting.cpp). А по 4-м парам точек можно найти персективное преобразование из одной лоскости в другую (плоскость поверхности земли в плоскость поверхности этого же участка Земли на кадре). Таким образом, мы можем любую точку на кадре, которая находится на поверхности дороги, перевести в географические координаты. Далее уже дело техники: для каждого объекта есть траектория в пикселях -> траектория в гео координатах -> считаем скорость. Считать скорость можно тоже по-разному, в ростейшем случае по формуле равномерного прмолинейного движения.
Но в реальности для получения точных результатов надо будет предпринять намного больше усилий, учесть дисторсию камеры, усложнить модель движения и т.д. и т.п.
Hello, I need the code for this project, can I have it please?
Yes, go to the video description
И как йоло4, пошустрее работает, чем третий?
Нет, но заметно точнее
hallo can i run this using google colab? can u guide me?
No, I don't use google colab. This project designed for servers and edge devices like Nvidia Jetson
do you have the code version python ??
No, only C++
🔥🔥
That's great ...
Knowledge coordinates are limited. Only the grid-based coordinate system using the geometric transformation using four points is accurate. You can translate the other videos and watch them. I also have a patent. Let's develop it together.
It seems like they were slower :p
what hardware did you use for this?
Laptop with Core i7 and GeForce 1060
Thank you very much.
How can I get the code 🙏
Link to the source code is in video description
Can we do it using python ?
You can, but I don't use it for realtime applucations
Hi Sergey I want to use this code to record the path and distance traveled by Mice in tests like this vid:ruclips.net/video/GebcshN4OdE/видео.html but in yolov4 since that video's actual software doesn't work as intended. I already have a trained model that works on different kinds of mice and is able to detect it on yolov4 using a collab notebook made by The AI guy.
Thanks :)
Did u get it ? I also want the same 😅
@@youtubeu4005 the github code in the link has been updated and works properly if you even want to ask the github creator a question he'll answer as well
@@mohamedfadl789 oohkk... But do u have colab code .. if u can share that it would be a great help... I don't have a gpu
@@youtubeu4005 Idk about the collab code thing, I can just give you this repo:github.com/colinlaney/animal-tracking it works the way you are referring to but from a computer vision approach, not a machine learning one, there's also this link to a supposedly working software that deals with these kinds of tests (have not tested it yet because I'm working on the code of the first link to make it work on the y-maze test): github.com/healthonrails/annolid other than those two links there's a youtube channel called the ai guy that deals mostly with yolov3/4
ps: if the 2nd link works please tell me how because I'm putting it on hold for now until I finish the y-maze
I have never working with an animal tracking. I think that it need a lot of work: videos, dataset, testing. I dont know how I can help you with this project