HUSKYLENS 2 Pose Recognition Function Description

1.Introduction to Pose Recognition

This function can detect the human body in the image, identify and plot 17 key body points, and then learn to recognize and track different human poses. It can detect multiple humans simultaneously, identify the human body across various angles, and predict occluded joint points to a certain extent. The 17 key points include the nose, eyes (both), ears (both), shoulders (both), elbows (both), wrists (both), hips (both), knees (both), and ankles (both), etc.

2.Pose Recognition Usage Guide

In this section, we will learn how to use HUSKYLENS 2 to detect human bodies in the frame and train and recognize specified poses.

2.1 Select Pose Recognition Function

Power on HUSKYLENS 2, and after it successfully starts, swipe the screen to find the "Pose Recognition" function.

2.2 Observe Human Detection Effect

Aim HUSKYLENS 2 at an image containing humans. Upon detecting humans, the screen will display white bounding boxes enclosing all humans in the scene, and 17 key points on the human bodies will be marked using dots.

2.3 Learning Pose, Observing Results

Position the human pose you want to learn, adjust the viewing angle of HUSKYLENS 2 so that the crosshair in the center of the screen is within the white box, then press Button-A on the top-right corner of HUSKYLENS 2 to learn this pose.

After learning a pose, when a learned pose is recognized, the screen will frame it with a colored box and display "name: IDx 90%" above it.
For example, "Pose: ID1 78%". Here, "name" defaults to "Pose" by default (customize the name via "Parameter Settings"). "ID1" refers to the first pose learned, and "78%" represents the confidence level, indicating the model's confidence in the recognized pose belonging to the learned set (e.g., "ID1 85%" means the model is 85% confident it is Pose ID1). The same logic applies to learning additional poses.

Multi-angle Learning: When learning a gesture, long-press Button-A to adjust the angle HUSKYLENS 2 "looks at" and perform multi-angle learning for the gesture. The learning progress will be shown during multi-angle learning.

3.Parameter Settings for Pose Recognition

HUSKYLENS 2's factory default parameters are sufficient for basic functionality. For more refined features, individual parameters can be adjusted manually. The following parameters are all based on the "Pose Recognition" function, so first ensure you have entered the "Pose Recognition" function, as shown in the figure.To select a parameter to modify, slide left or right on the parameter text below the screen.

3.1 Forgetting IDs

To forget all previously learned poses: Step 1, tap "Forget ID" on the screen; Step 2: a pop-up window titled "Forget all IDs and names" appears; then tap "Yes".At this point, aim HUSKYLENS 2 at the previously recorded pose. A white bounding box will appear on the HUSKYLENS 2 screen, and if the pose ID is not recognized, it indicates the "forget" operation is complete.

3.2 Detection Threshold

The detection threshold controls the sensitivity of human detection: a lower threshold means a looser standard for classifying an object as human (more likely to mistake non-humans as humans but rarely misses real humans), while a higher threshold means a stricter standard (less likely to misjudge but may miss real humans).

Setup steps: Tap "Detection Threshold". A parameter adjustment slider will appear above it. Slide left to decrease, right to increase. The effect is illustrated in the figure.

3.3 Recognition Threshold

Recognition Threshold controls the strictness of pose recognition: A lower threshold makes it easier for the system to match a detected posture to a learned pose (more prone to false recognition but less likely to miss recognition), while a higher threshold makes the condition stricter (less likely to miss recognition but more prone to false negatives).

Setting Steps: Tap "Recognition Threshold". A parameter adjustment slider will appear above it. Sliding left decreases the value, sliding right increases it. The effect is shown in the figure.

3.4 NMS Threshold

NMS Threshold is a common parameter in visual recognition used to filter detection bounding boxes. In visual recognition tasks, models often predict multiple overlapping bounding boxes around the same target. Without filtering, a single object may be boxed by multiple overlapping boxes. Adjusting the NMS threshold can remove these redundant boxes and retain only the optimal one.

In simple terms, the NMS threshold determines how much overlap between two boxes is considered "redundant". For example, a low threshold (e.g., 0.3) will mark two boxes as redundant if they have any overlap, removing one. A high threshold (e.g., 0.7) requires significant overlap to be considered redundant, potentially retaining more boxes.

A higher threshold works well in dense/occluded scenarios, capturing more humans but possibly resulting in "multiple boxes per person". A lower threshold is suitable for clear single-person scenarios, ensuring one box per person but potentially missing blurry figures.

Setup steps: Tap "NMS Threshold" to display a parameter adjustment slider above it. Sliding left reduces the threshold, resulting in fewer detected people; sliding right increases the threshold, resulting in more detected people.

3.5 Set Name

This parameter allows you to set a name for a learned pose, which can be in both Chinese and English.

This parameter allows you to set names for the learned poses, in both Chinese and English. Setup steps: Tap "Set Name"; slide up/down the number in the top-left corner to select which ID to name; tap the keyboard on the screen to enter a name (as shown in the left image); once set, tap the √ icon in the bottom-right corner to save. Successfully saved, a green checkmark will appear in the top-right corner.

3.6 Display Name

This parameter controls whether to display the name when a pose is recognized. The default is to display the name.

Setting Steps: Click "Display Name". The switch above it turns blue, indicating it is enabled. When a learned pose is recognized, its name will be displayed (refer to the left image). Click the switch, and it will turn white, indicating it is disabled. When a learned pose is recognized, no name will be displayed (refer to the right image).

3.7 Restore Default

This parameter restores all settings to their default values, forgets all learned IDs and names, but does not clear exported models (see "Exporting Models" below for details).

Setting Steps: Click "Restore Default", and after the "Restore Default Configuration" dialog appears, click "Yes".

3.8 Export Model

This parameter allows saving and exporting the current set parameters, learned IDs, and settings to the local memory of HUSKYLENS 2. It is applicable for scenarios such as migrating parameters to another HUSKYLENS 2. This operation does not require a TF card.

Export Steps:
Click "Export Model". When the "Save Configuration To" pop-up (as shown in the left figure) appears, scroll to select which model to save (up to 5 models can be stored). Next, click the "Yers" button in the lower-left corner of the pop-up to save. After confirmation, the export process will be automatic, as shown in the right figure.

View Exported Model: After the "Exporting" pop-up window disappears, you can view the exported model file on your computer.

First, connect the HUSKYLENS 2 to your computer's USB port.

Next, you can access the memory of HUSKYLENS 2 through a computer via the path shown in the figure below. Here, you will find two model-related files with extensions .json and .bin. The numbers before the extensions indicate the "model number" selected when saving the configuration. Both files can be copied and pasted to other locations.

3.9 Import Model

This parameter allows importing a pre-trained model from HUSKYLENS 2 (referred to as "HUSKYLENS A") to another HUSKYLENS 2 (referred to as "HUSKYLENS B"), enabling HUSKYLENS B to replicate the learned postures and adjusted parameters of HUSKYLENS A without reconfiguring parameters or retraining.

Import Steps:
Step 1: Connect HUSKYLENS A to a computer and copy the exported model file to the computer desktop.
Step 2: Connect HUSKYLENS B to the computer and paste the file from the previous step into the default specified folder of HUSKYLENS B (path shown in the figure). (If the pose-recognition folder is not found, first manually create the folder, as the folder will automatically be created after importing a model. Then return to Step 2.)

Step 3: First, confirm that you have entered the "Pose Recognition" function, then click "Import Model". After the "Load Configuration" pop-up appears, slide the number selector up and down to select the model number. This should match the model number saved in the prior step.
For example, if the model file pasted into HUSKYLENS 2 is config1.json, select number 1. Finally, click "Confirm" to import.Wait for the "Loading" pop-up to disappear, and the import is now completed.

Check whether the parameters and learned postures of HUSKYLENS B and HUSKYLENS A are consistent. The left image below shows the recognition status of HUSKYLENS B before model import, and the right image shows the recognition status after import.