Human Keypoint Detection[AI]

## Human Keypoint Detection[AI]

By running a trained model, it classifies and locates the targets that need to be identified.

## Module Principle

Construct the relevant input, output and image display of the module by running inference on a trained model to obtain the coordinates, bounding boxes, confidence scores and other information of human body keypoints.

### Algorithm Workflow

Run inference on Google's pre-trained model to obtain the coordinates of hand keypoints.

### Algorithm Formulas
Anchor-free Decoding: Directly predicts box coordinates via sigmoid(tx/ty) and exp(tw/th), without predefined Anchors.
Total Loss Composition: L total = λ cls * L cls + λ reg * L reg + λ obj * L obj, where regression loss uses CIoU to improve box localization accuracy.
SimOTA Assignment: Dynamically selects positive samples through a cost matrix C i,j, replacing traditional Anchor matching to improve the rationality of sample assignment.

## Usage Scenarios

Contactless human-computer interaction for smart devices
Accessibility assistance and sign language digitization
Industrial intelligent manufacturing and human-robot collaboration
Security monitoring

## Usage Method

### Predecessor/Successor Modules
Camera Color Image, Video Read, Label Filter, Display Image, etc.

### Usage Example

![](/media/202602/2026-02-11_165752_3961080.3865130323940573.png)

## Parameter Configuration Description

![](/media/202602/2026-02-11_152807_6719900.5611970532113871.png)

## Output Description

![](/media/202602/2026-02-11_152829_5345190.682360651024562.png)

## Function Attribute Description

![](/media/202602/2026-02-11_152857_9528420.6948676270515776.png)

![](/media/202602/2026-02-11_152944_5101780.3116355092483978.png)

---

Click here to jump to the corresponding Chinese manual page: [Human Keypoint Detection[AI]](/doc/327)