Real-Time 3D Pose Detection & Pose Classification | Mediapipe | OpenCV | Python
HTML-код
- Опубликовано: 16 июл 2024
- In this post, I will show you step by step how to do real-time 3D pose detection/estimation in python using mediapipe. Later on, we'll combine this with some heuristics to create a pose classification system.
Timestamps:
00:00 Intro
00:53 Pose Detection & Mediapipe's Implementation
06:44 Pose Detection in images and on videos
29:54 Creating Pose Classification System
Join Computer Vision Course: bleedaiacademy.com/computer-v...
Blogpost: bleedaiacademy.com/introducti...
Download Code: app.monstercampaigns.com/c/ka...
Get 1 on 1 AI Coaching with Taha: pensight.com/x/tahaatbleedai
Support Bleed AI on Patreon: / bleedai
If you want us to create AI and computer vision solutions for you,
then reach out to us here: bleedai.com/services/
----------------------------------------------------------------------------------------------------------------------------------------------------------
If you enjoyed this video then don't forget to subscribe and like this video.
🎇 Subscribe to this Channel: bit.ly/Bleedsub
🎇 Subscribe to Bleed AI Blog: bleedaiacademy.com/subscribe/
🎇Check out our Courses: bleedaiacademy.com/courses/
Follow Bleed AI for more 👣 :
◼ Website: bleedaiacademy.com/
◼ Facebook Page: / bleedaiacademyy
◼ Linkedin: / bleedaiacademy
◼ Instagram: / bleedaiacademy
◼ Twitter: / bleedaiacademy
◼ Patreon: / bleedai
#mediapipe #computervison #OpenCV #posedetection #poseclassification #poseestimation
You just saved my life with this video, totally outstanding explanation!!!
Thank you so much for you time and effort in preparing such detailed comments on the code notebook, it is extremely helpful. Looking forward for the course.
Thanks a lot, Ruthvik for your kind words, it means a lot.
Always Love your Videos, You are one of my favorite teacher of Computer Vision.
Best deep learning tutorials ever. Thankyou so much for such detailed explanation
You're very welcome!
Thanks alot, you really have the ability to explain it. Great video!
Great work BleedAI. Its really helpful
Thank you Rafay :)
Beautiful work.
Thank you
Thank you so much s for your nice explanation. con you help me to find the accuracy of each pose classification. looking forward to the upcoming courses.
Very well done. Thank you.
Amazing!
Fantastic job helped to build my own project
Thanks
Thanks a lot bro. awsome video.
Glad you liked it ;)
It's a very useful project. I see that you also talk about the 3D camera, did you ever run this project with 3D camera? And I have a question that how can I convert the depth get from the camea into pixel for easier to calculate angle? Thanks for your video.
Thank you so mucch...
Awesome
Thank you
thankyou
it was a really great video tutorial.
thank you so much for your time.
looking forward to new videos
Thank you.
Thanks for your video. Could you please tell me how can we remove irrelevant markers from our detection? For example, I only need the main joint markers and do not need the face and fingers.
Thank you very much for making this tutorial. Please what are some possible applications of this concept of real time pose detection and classification in say industrial systems or robots?
Hi, You're welcome and I do mention some industrial and other use cases at ruclips.net/video/aySurynUNAw/видео.html Also It's being heavily used in sports analytics.
Thank you very much Sir
good
could you please tell me how to convert into web application using flask.which part of the code we have to include?
aarigatooo
Hi. Do you know how can i load 3d animated model/character into python and make that model/character work with pose detection?
Hi,
Thanksfor your time and effort. in time 29:20..I'm using ur exact code and I don't see any feedback when I run the code.
in your video it shows your pose detection but I dont see any window with a video of myself that is opened.
What am i missing?
hello can you make person tracking with mmpose , so that their id will not swap even they cross each other
pose_video = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.7,
min_tracking_confidence=0.7)
this particular line of code is giving name error what to do?
Hello sir, i have a doubt, loved this video but i am trying to make an action recognition model, so can i used the landmarks which are basically x,y,z, coordinates to train my model? Will it be efficient. Please reply
Hi, can we do the same thing with image instead of video?
Can you please provide code with explanation video for multi pose estimation
the kernel has died problem in the 3rd section when defining the result variable
Please can u tell me name of the algorithm of mediapipe that you use for human detection and classification? please reply as soon as u possible.
I really wanted to see the 3d pose render on a video
Person detector will detect and crop the object and before it is sent to tracter for tracking the motion of Object first it is send to landmarks detector I'm i right
Hello I want to train this project for multiple classes make it. For example dancing running play footbal .. How can i make and how can i train for
hi im doing a cricket pose-resurrection project for my FYP
awesome video. can we save the cordinates of the live feed into a csv file
Yes, you can save pose coords live on a CSV file using a file writer in python
How to do the pose classification using keypoints in neural network?I mean , do we have any pre-build neural network for mediapipe which uses those keypoints for classification? Can you please guide me thoroughly this? Also How to make the classfier that make predictions on an unseen camera angle? Also I am some graph error after calculating keypoints of 3000-4000 images in a loop .
I want the best accuracy possible.
Thank You
Hi Ayush, please take a look at how you can train a DL model on keypoint coordinates online, you can then use the techniques for this use case. Also, there is no easy way to predict on unseen angles, the best way is to record the exercise at different angles and then the DL model will be more robust at detecting those exercises during prediction time.
Is it possible to run mediapipe pose estimation with GPU for better performance?
Hi can you make a tutorial on how to make it for the pictures, still images ? not live. I need a tool to classificate images by pose on a website :D
Can it run in colab
Hello, thank you for the tutorial on this, however, the source code is missing. Seems like it was deleted, kindly check on that
I am trying to fine tune this tflite model on some exercises , but i can not deal with the model as tflite and can not convert it to keras model. what should i do?
Hi salah, you can convert the model into ONNX format using tflite2onnx and then import it into a framework in which you are comfortable in.
can you provide research paper of this project
Hi, when I run the setup pose function for video, it says the 'Your session crashed. Automatically restarting'. Could you please let me know what could be the bug?
did you find the bug??
hi sir, every time i try to run the code for Pose Detection On Real-Time Webcam Feed/Video i got error as "Your session crashed for an unknown reason" working on google colab
,plz help!!
Exactly. I am too getting the same issue in Jupyter notebook. Please help
Is there a good resource that can help me how to get started with training this model on specific exercises?
Not exactly, but the process is typically the same for all exercises, you feed landmarks into a dense NN or use heuristics to determine the exercise. We also have an exercise-based demo coming up next week.
How to remove face points and keep others
sir i'm getting error while installing mediapipe, plz help me with this
Thank you very much for your sharing. Can I repost it on Chinese video websites? I am a college student from China. Thanks again for the video!!👍👍👍
Sure, you can share as long as you give credit.
ERROR FOR 3D Plot Pose landmarks: AttributeError: module 'mediapipe.python.solutions.drawing_utils' has no attribute 'plot_landmarks'. How fixed this issues?
Hi MorsaliN, please make sure that you have installed the latest version of the mediapipe.
i am not able to get the code, could someone please help me
sir, can i use this pose estimation code to make a squat exercise classfication analysis???, i mean like whether or not the squat is deep enough, the back rounding, butt coming up earlier etc?. i tried to use YOLO model and collect all the dataset needed however it was such a bad method and the model was so inaccurate. and do you have any advice for my case??
Actually, I do squat detection use pose model in our course called. "Computer Vision For Building Cutting-Edge Applications"
What's the mediape version? It's 0.8.3, but it says there's no option for model_complexity.
I think this option was removed in later versions but would need to confirm this.
Nice explanation btw from where did you buy naruto action figure please share the link ... to buy and also i am from india.
Thanks, Takshil, I got the figure from aliexpress website, don't have the exact link.
Assalam o alaikum, I new into the subject and I don't know python too. kindly guide me how to do some projects.
Walaikum Asalam ibrar, you can check out my resource guide: bleedai.com/24171-2/
thnks so much you are professional but i have a problem with :# Setup Pose function for video.
pose_video = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.5, model_complexity=1)
NameError: name 'mp_pose' is not defined
yes sire me too
Hi, make sure that you have executed all the cells in the jupyter notebook, the mp_pose variable was initialized in the second code cell.
Is it achievable in javascript and not a pain because I had to work a lot just to get the pose detection working in my react app
The code is pretty much the same for mediappe JS bindings minus some syntactical changes. Check mediapipe website, for JS examples
how to select the ROI from the key points(landmarks) generated skeleton of human..??
If you just want to draw the bounding boxes around the person, then instead of using draw_landmarks using draw utility module, use draw_detection. reference: github.com/google/mediapipe/blob/master/mediapipe/python/solutions/drawing_utils.py
Also, if you want to extract the box ROI, then look for an attribute like `box`, in the results object, btw I don't remember finding one, even though there should have been one present.
Now the best way to grab the box roi, is to draw box around the extreme coordinate points, I'll do exactly this in an upcoming tutorial 2 weeks from now. for now take a look at this rough solution: stackoverflow.com/questions/66876906/create-a-rectangle-around-all-the-points-returned-from-mediapipe-hand-landmark-d
Thank you
Hello friend, I hope you are very well. I have the following questions and I hope you can help me.
how should I interpret a negative z value? For example -0.57. Does this mean that marker is much closer to the camera? The documentation says that the smaller the z value, the closer that marker is to the camera.
Now when plotting using the mp_drawing.plot_landmarks and pose_world_landmarks functions, the resulting graph will be expressed in meters but originating from the center of the person's hips. How can I get the distance in meters from each marker to the camera? This is possible?
thank you very much, greetings!
You won't be able to get the distance from one camera
And the negative value is taken from the fact that a cube with sides from 1 to -1 is drawn around the object and a lot of things can be learned and coded from them
@@oleggerbylev4069 hi man. Thanks for yours anwers.
I'm working on a project where I want to detect where people are looking (headpose estimation). I have done it for single person but not able to do for multiple people. I want to store landmarks of multiple people so that I can do headpose estimation on each person individually but I don't know how to store these landmarks for multiple people. I'm using mediapipe and opencv. Would you please help me sir in doing it??
Hi, you need to increase the num_face variable to 2 or more when you initialize the face landmark detection function in mediapipe, I think the default value is 1, which is why you're not able to detect more faces.
Can you import mediapipe in vsc, and run this code in vsc?
Yes
🙊💓
Hi Sir , can we have video of multiple person tracking under multiple cameras using mediapipe
Hi Siddharth, I got your other messages from the course, basically you need to find a good object tracking solution that is robust to occulusion, try Deepsort tracker or it's variations with a detector.
hOW TO POSE DETECT 2 PEOPLE AT THE SAME TIME?
are landmarks of a person same in two videos recorded from different angles?
No landmarks x,y coordinates depend upon the camera angle.
Hi Sir, I am developing an android application using flutter which needs to use this python script is there any way in which we can integrate this with flutter.
You can embed a Unity module in flutter.
the unity module will be using mediapipe's C# API, we did something similar for one of our clients at Bleed AI Solutions.
pls telll using it with 3d models ls
I'm not sure I understand what you meant
Hi Sir, can we do pose estimation for multiple person using mediapipe??
Mediapipe is not designed for multi person
Hello bro its seems like the drive link for the source code is not available.could you share the source code .
drive.google.com/drive/folders/1xpKpajJ85tS2iNSxCHBC1dCo2Z935H0F?usp=sharing
codes are not available ,can u reupload that?
Hello , I got an email for subscription and after I did that I got a link to the source code but it is not available inside the link.
I meet the same problem, have you solved this?
Oh! I find three weeks ago vishwarajasekar6319 asked this question and got the answer!😄And the link is still available! drive.google.com/drive/folders/1xpKpajJ85tS2iNSxCHBC1dCo2Z935H0F
Hello Sir, how can i make it detect landmarks on multiple people.. Please help
Hi, this particular model is only able to detect poses for a single person in a frame, you can combine it with person detection for multiple people or alternatively use a multi-pose detection model.
If you had done on multi people detection please help me with the code...
Please in the website you mention i cant see it ..The website you displaced ...
Hi, Benya not sure what you mean
@@BleedAIAcademy the mediapipe course link is not working and the code link is also not accessible mentioned in the description
@@Krishana_007 You need to open this link in a browser where adblock is disabled: ruclips.net/user/redirect?event=video_description&redir_token=QUFFLUhqbE5BOHNPMzVuOEFhcjJ1NWsxTUFkYWw5VkR0QXxBQ3Jtc0tsZFVSOTZ1Q3lhNS1sNUNUYmlpMTlLemdxS2FNRVc1aUx6LVNZb2hGcVZkak5fU0FFQ2xVZHFVQnlGc1g1YXZaZ1BjNGp1MGRtal9tT0hic0N0MC1TUHhoVU1wa3pUYjlHNmVDdmgzZXR0Z0t6RmtfNA&q=https%3A%2F%2Fapp.monstercampaigns.com%2Fc%2Fkalfyxphljhqu1zouums%2F
Also the mediapipe course sign up link: app.monstercampaigns.com/c/gbpbgnzbsex4ynffbs5j/
why me eror in pose detection on real time webcam
Hi, what error are you facing?
How can i contact you ,please help me out, its my final project
You can book a meeting with me here: pensight.com/x/tahaatbleedai
Getting Error :
line 43
if (time2 - time1) > 0:
^
SyntaxError: invalid syntax
⚠⚠⚠⚠⚠⚠⚠
can you please give me the code your website is not working
app.monstercampaigns.com/c/kalfyxphljhqu1zouums/ You can download the code from here. Make sure to disable adblock
三哥这口音真无法吐槽。。。😂😂
How to download the code
Hi, download link is present in youtube description
@@BleedAIAcademy yeah..but it couldn't download for me..
No code available ..
Code Download Link: drive.google.com/drive/folders/1xpKpajJ85tS2iNSxCHBC1dCo2Z935H0F?usp=sharing
@@BleedAIAcademy Salute ! Thank You for Giving Me Your Valuable Time in very Busy life
Speak much more slowly, please. English with an Indian accent is hard.