Artificial Intelligence-Driven Digital Interactive Design And Public Communication Of Hua Tuo's Five-Animal Frolics
Feb 26, 2025
3. Motion Capture Based on Artificial Intelligence Algorithms
To enable the interactive functionality of the Five-Animal Exercises device in front of a camera, the core lies in establishing a connection between virtual and real-world representations. The participant exists in the real world, but must first be reconstructed in the virtual world. Then, the virtual character is matched to the reconstructed image of the participant, enabling various interactive features. To achieve highly synchronized interactions, two major technical challenges must be addressed: Human Pose Estimation and Human Motion Retargeting. Traditional human reconstruction methods often rely on expensive motion capture equipment, but this study leverages the advantages of artificial intelligence to achieve human reconstruction at a lower cost.
New TCM Herbal Cistanche for bodybuilding
The primary goal of human pose estimation technology is to detect human positions and keypoints from video streams. This technology can be applied to two key features of the device design:
Real-time video streams are captured via a camera, and the model detects human keypoints, preparing for subsequent operations.
To meet the requirements of the Five-Animal Exercises movements, video clips of these exercises are preloaded into the model, which detects the keypoints for each movement and generates skeletal sequences of the Five-Animal Exercises characters, preparing for the transfer of user representations.
New TCM Herbal Cistanche For Body building
The following section provides a detailed explanation of the technical implementation process.
New TCM Herbal Cistanche For Body building
3.1 Skeleton Extraction
This study utilizes the OpenPose tool [13] to achieve real-time 2D keypoint capture and recording. OpenPose is primarily used for real-time multi-person 2D pose estimation. By detecting and associating the keypoints (such as joint positions) of each person in an image, it enables pose recognition, though it does not provide 3D surface information or detailed segmentation of body parts. This method introduces the concept of Part Affinity Fields (PAFs), which, through convolutional neural networks (CNNs), simultaneously predicts confidence maps for body parts and affinity fields between parts. Keypoints are parsed using a greedy inference process, offering strong real-time performance.
With the pre-trained OpenPose model prepared, this study feeds video stream data (using demonstration videos by national champion Cao Jin) into the model for inference, extracting skeletal data to obtain skeletal sequences (see Figure 4).
Fig.4 Detection Results of Human Body Keypoints
3.2 Dense Human Pose Estimation
2D skeleton extraction provides limited support for subsequent motion retargeting tasks and is only applicable to simple character mapping onto 2D skeletons. To meet the needs of more complex character representations, the 3D surface structure of the human body must also be extracted. DensePose [14] aims to achieve dense human pose estimation by mapping all human pixels in an image to their corresponding points on a 3D human surface. This method not only detects keypoints but also provides detailed surface parameterization for each body part, outputting UV coordinates and body part segmentation masks.
The DensePose-RCNN system utilizes a convolutional neural network (CNN) that combines the DenseReg method with the Mask-RCNN architecture. It employs a cascade of region proposal generation and feature pooling, then uses a fully convolutional network for dense predictions. This generates discrete part labels and continuous surface coordinates, predicting the dense correspondence relationship of the human surface through training.
Since DensePose provides rich body surface information, it is suitable for applications requiring detailed human shapes and appearances, such as 3D modeling, virtual reality (VR), and augmented reality (AR). Ultimately, DensePose is capable of generating detailed data that includes human surface information and motion sequences, supporting the transfer of more complex character representations (see Figure 5).
Fig.5 Hua Tuo's Five-Animal Frolics Video Image and Dense Pose Comparison Chart
3.3 Human Motion Retargeting
Whether enabling a virtual character to mimic a user's real-time movements or animating a user's photo to perform the Five-Animal Exercises, it all relies on Human Motion Retargeting technology. This technology redirects a source motion sequence to a new motion sequence. There are various approaches to achieving this, and based on the requirements of the interactive device, this study adopts two applicable techniques.
For the extracted skeleton sequence, simple processing based on the skeleton is sufficient to transfer it to a new form. For instance, by designing a "stick figure" model, we can enlarge the head keypoints and add different colors to the body parts, thereby generating a new representation. Table 3 illustrates the effects of motion retargeting using the Five-Animal Exercises video.
Tab.3 Results Containing Key Point Image and Image From Retargeted Video
For more complex motion retargeting scenarios, diffusion models can be introduced to assist in achieving the goal. The MagicAnimate framework [15] allows for generating video animations that adhere to a given motion sequence while maintaining temporal consistency and fidelity to the reference image. This model embeds the reference image into an appearance encoder to generate an appearance embedding, while the target pose sequence (e.g., DensePose) is input into a pose control network (ControlNet) to extract motion conditions. Based on these two signals, the model trains a video diffusion model that animates the reference identity according to the given pose.
Using this feature, our interactive device can take the user's motion sequence as the target pose sequence and the Five-Animal Exercises character image as the reference image. This enables the user's free movements in front of the camera to drive the synchronized motion of the Five-Animal Exercises character, creating an interactive experience (see Figure 6).
Fig.6 Screen Effect Generation
4. "Move, Stay Healthy, and Enjoy" Public Interactive Device Design
4.1 Functional Features
The "Move, Stay Healthy, and Enjoy" interactive device leverages cameras and artificial intelligence technology to enable participants to perform a series of tasks in front of the device. These movements are then recorded in real time and matched with virtual character images, providing an immersive interactive experience. The application scenarios of the interactive device are shown in Figure 7.
Feature 1: Users can choose from a variety of virtual character representations based on personal preferences, including lively kung fu characters, uniquely styled block characters, and simplified stick figures. Through these visual effects, users enjoy a rich interactive experience. These characters not only mimic the users' movements but also respond to them.
Feature 2: Users can take full-body photos in front of the camera to enhance the experience of performing the Five-Animal Exercises. The system inserts the user's image into the Five-Animal Exercise video clips, allowing users to imagine themselves as masters of Hua Tuo's Five-Animal Exercises, executing various professional movements with precision.
Fig.7 WILDFIT Application Scenarios
4.2 Parameter Settings
The device is named "Move, Stay Healthy, and Enjoy" to establish a connection between animals, nature, and fitness. The goal is to make the design a bridge between these three elements. Based on this concept, the device is an interactive screen installation intended to be placed on outdoor lawns or in amusement parks. It is equipped with various parameters, as shown in Table 4.
Table 4: "Move, Stay Healthy, and Enjoy" Parameter Configuration
| Parameter | Details |
|---|---|
| Screen Size | 2.5m × 1.5m |
| Resolution | 1920 pixels × 1080 pixels |
| Interactive Buttons | Start button; motion tracking and feedback indicator button; exit button |
| Touch Sensitivity | 10ms response time, 30FPS frame rate; supports hand gestures, body posture, and physical sensing |
| Feedback Interaction | Voice prompts; or visual feedback (e.g., indicators showing unaligned body posture or incomplete motion gestures) |
| Durability | IP65 waterproof rating; aluminum alloy material shell
|
Fig.8 WILDFIT Design Concept Demonstration
4.3.2 Physical Evaluation: Contents and Methods
To further assess the impact of the physical device after its application, the following evaluation methods are proposed:
1) Physical Health Indicators
Purpose: Compare participants' physical health indicators before and after engaging in the fitness program (e.g., weight, body fat percentage, heart rate, blood pressure) to evaluate the fitness program's impact on physical health.
Method: Linear Regression Analysis
Linear regression is used to analyze the changes in participants' physical health indicators before and after the fitness program. The formula is as follows:
Y = β₀ + β₁X + ε
Y: Dependent variable (e.g., weight change).
X: Independent variable.
β₀ and β₁: Regression coefficients.
ε: Random error.
2) Psychological Health Indicators
Purpose: Investigate participants' psychological health indicators after participating in the fitness program (e.g., emotional state, stress levels, happiness) to evaluate its impact on mental well-being.
Method: Logistic Regression Analysis
Logistic regression is used to evaluate the program's impact on psychological health indicators (e.g., happiness). The formula is as follows:
P(Y=1) = 1 / (1 + e^-(β₀ + β₁X))
Y: Binary variable (e.g., whether the participant feels happy).
X: Independent variable.
β₀ and β₁: Regression coefficients.
For the "Move, Stay Healthy, and Enjoy" device, the independent variables are detailed in Table 5.
3) Engagement and Satisfaction Survey
Purpose: Investigate participants' engagement and satisfaction with the fitness program.
Method: Design and distribute a survey questionnaire to collect user feedback on the device's usability, enjoyment, and practicality, as well as overall satisfaction with the Five-Animal Exercise interactive device.
Participants will rate these aspects, and the feedback scores and suggestions will be used to evaluate the device's effectiveness.
