Real-time Body Pose Estimation A Program That Detects With The Support Of Artificial Intelligence And Design
1.1 Purpose of the Project
Program and design project that detects real-time body pose estimation with artificial intelligence support
in many fields of computer vision, such as human-computer interaction, artificial intelligence and sports.
is an applicable research topic. The main objective of the project is health monitoring with Artificial Intelligence
real-time estimation of 3D human body poses for use in applications
and calculates that the movements are made at the right angle and provides the user with a program that
about posture. It helps users to maintain the correct body position while practicing sports.
to improve their experience by supporting them to protect themselves, enhance their sports performance and
providing AI-powered proctoring to make their workouts more efficient is key
is one of the objectives. In line with this objective, the real-time data received at the beginning of the project
body poses are detected from the images and this information is analyzed by a deep learning based model.
with the help of deep learning and machine learning algorithms. Using Deep learning and Machine learning algorithms,
The first step, the system model, was developed for predicting body poses. This
system model was applied to the dataset and achieved high performance. This project
With the aim of providing users with a more effective sports experience, we are working with technology and artificial intelligence
coaching and analyzing the user's body form in real time.
aims to provide suggestions for correction and improvement. Up-to-date with continuous development
working on the research topic and gaining different perspectives on development are also among the objectives.
is one of them.
Real-time body pose estimation has been a major area of research in artificial intelligence and image processing in recent years.
development. Similar examples in this field include many projects around the world and in our country.
covers. Research has shown that various projects developed in this field, especially artificial intelligence
clearly demonstrates that it includes the field of assisted body pose estimation. Worldwide
OpenPose, Google MediaPipe, Microsoft Azure Kinect platforms in similar projects
has been used. Pose estimation has been studied in academia and R&D projects.
Ahmet Samet HALICI and Ayşe DEMİRHAN's [1] Kalman filter and global nearest neighbor
multi-person real-time pose tracking project with the multi-person real-time pose tracking method
are cited as examples of such practices. Such projects are often used in the field of health, mobility
analysis applications, the game development industry and virtual reality applications.
are used. One of the pose analysis methods is machine learning based methods,
Considering the studies in the literature, maximum accuracy compared to other methods
has been observed.
There are additional indicators and features that distinguish this project from other applications.
A software that instantly returns the body position value and warns if the posture position is incorrect
and all optimizations have been enhanced by making it easier for users to read the expressions.
included.
Computer vision modeling with human pose estimation, such as limbs or joints in the human body
to try to identify the important points and to determine one's position in the moment.
is a project that can help. It is a project where movement patterns are analyzed and based on the input
predictions are made that decisions will be made in the forward direction. Detecting body parts and
smartphone camera or surveillance lens to track them in motion in 3D space
RGB image with an optical sensor such as an optical sensor. Body pose estimation,
detects points on the human body and identifies the points as X, Y and Z of the joint position.
in a space where it is represented on axes. The most basic form of human pose estimation 2
is dimensional point extraction. This is a depth perception method that only takes into account 2-dimensional space and
is not, it usually requires a model with minimum accuracy. More common
the approach used is more efficient in tracking, given favorable lighting conditions.
are accurate 3D pose estimation models. Both methods are often developed together
used because the 2D approach is faster in detecting real key points, while the 3D
models provide support for accurate perception of space.
Research has found that three main methods are used for human pose estimation. The first method is
Skeleton-based pose estimation model identifies a person's skeleton from an image or video. Human
body parts on the skeleton identified based on the previous model of the skeleton
classify and describe the orientation of a series of joints and limbs in a person's body.
is retrieved. The second method, Contour-based models, is to represent the human body as a series of contours.
and represents the connection of body parts, which is difficult in skeleton-based models.
capture. These models are used for 2D pose estimation. As a final method,
The volume-based model is more advanced than previous methods and can be used to represent the human body in a 3-dimensional
as volume. A realistic representation of body poses in geometric shapes such as cylinders and cones.
shapes. The main advantage of 3D pose estimation is the wide range of motion.
more complex body positions can be modeled by capturing them. However, this approach is too
time-consuming and consumes more processing power than 2D pose estimation.
The real-time body pose estimation project is a real-time pose estimation project that includes image processing and artificial intelligence algorithms.
consists of a series of steps. The first step is to detect human body positions in the snapshot.
artificial intelligence algorithms were used. Once the body positions are detected, the pose
analysis model developed using convolutional neural networks to determine the state of
is used. This model is trained with deep learning techniques and the data categorized into pose categories
cluster. Finally, the identified pose states are used to perform instantaneous analyses and
to present it to the user.
As a result of the improvements and feature additions made throughout the project scope, the success of the revamped model
With the help of the graphs drawn, the rate and usability increase in direct proportion.
observed. Commonly used combinations for model training and live images
important in the process of identifying points of the human body, such as limbs or joints
mathematical functions have been tested in studies and the classification algorithms
the impact on the performance results is analyzed in detail. By making evaluations, the optimization
successful, strongly performing classification algorithm and real-time convolutional neural
networks architecture to classify body positions processed in real time.
project that can perform its task simultaneously.
Pose detection and position sensing systems in combination with other computer vision techniques
health, coaching, commercial, military, security, social, psychological, pedestrian's body in autonomous vehicles
language detection, recognition of hand movements, limb rehabilitation training, sports guidance,
surveillance and even virtual rooms. These private spaces
to facilitate human body pose estimation. Real-time detection of human body pose estimation,
will lead to significant advances in the field of computer vision, with wide application use
has potential. This technology can provide more accurate, reliable and cost-effective results.
research and development activities are ongoing to produce the system.
#4.1 Technologies, Platforms and Languages Used in the Project
Development of the project application, body position prediction with artificial intelligence and biceps curl
MediaPipe framework and Python programming language to create a (sports movement) counter
to predict body joints and parts. In the project, MediaPipe is used to predict body joints and parts.
used by the AI models. This framework takes the image or video input and uses artificial intelligence models
by detecting objects in the input and providing successful outputs.
The project provides wide framework and library support, easy readability, artificial intelligence
models and image processing processes can be effectively implemented in Python.
programming language. Creation of the model, installation of libraries, plugin
support and interface connections were written in Python language version 3.11.
version of Python.
For image processing, we used OpenCV (Open Source Computer Vision Library).
advanced Python libraries were used in the project. OpenCV, image processing and video analysis
is a library that provides a set of functions and tools. In the project, taking images from a computer camera,
OpenCV is used to visualize and process the image. This
technologies provide solutions in image processing and artificial intelligence applications.
Trigonometric operations, a subfield of Mathematics, to calculate angles between body joints
is used. The math library is set up and trigonometry functions are used
the calculation of the angles between the joints is facilitated. The development experiments were designed on Google Colaboratory and Visual Studio Code platforms.
Colaboratory enables fast training of deep learning and machine learning models on GPU.
results can be accessed and developed, and no additional downloads are required due to ready-made Python libraries.
is a free cloud service tool that allows easy use without the need to be heard.
The Visual Studio Code platform allows the model to be tested and the poses in the snapshot
to provide connectivity to the space where it will be shown. These technologies and platforms are
for AI pose prediction and biceps curl (sports movement) tracking
is used. By combining MediaPipe, Python, OpenCV and trigonometry features,
facilitates the processing of images from the camera. Predicting body joints
motion tracking by calculating angles, calculating angles and estimating posture using this information.
with accuracy.
%pip install mediapipe opencv-python 
)
)
)