Use webcam to detect and visualize hands (3D World space)
An example on how to get camera feed, detect and visualize hands in real time in 3D world space.
Last updated
An example on how to get camera feed, detect and visualize hands in real time in 3D world space.
Last updated
This example demonstrates how to load and display camera feed in a Unity scene with a and an , implement hand tracking with the , and use the to render detected fingers in 3D world space.
Open the Unity Project you created in the section.
Right-click on the Assets
folder and select Create > Scene
.
Type the scene's name. In this example, we'll use the name WebcamDemo3D
.
After the scene is created, right-click on the scene and select GameObject > UI > Canvas
.
Navigate to the LightBuzz Prefabs
folder at Assets\LightBuzz Hand Tracking\Runtime\Prefabs
.
Then, right-click on the Hierarchy
pane and select Create Empty
.
Give a name to the new component. In this example, we'll use the name Demo
.
Then, go to the Inspector
pane and select Add Component
. In the search bar, type new
and select the New script
option.
Type the script's name and select Create and Add
. For this example, we'll use the name WebcamDemo3D
.
Double-click on the newly created MonoBehaviour
script and import the necessary namespaces.
After adding the serialized fields, go to the Unity Editor to connect these fields with the Demo
component.
At the Inspector
pane, select the round button next to each SerializeField
.
Then, at the Scene
tab, select the corresponding prefab. For example, for the Image
field, select the ImageView
prefab.
When all fields are connected, the result should resemble the following image.
Then, select the HandManager
prefab. Connect the Image
field to the ImageView
prefab and uncheck the Is 2D
option.
Make sure the Is 2D
option is NOT selected to see the hand tracking detections in the 3D world space. If the option is checked, detections are displayed in the 2D space.
After connecting all the fields, navigate to the Canvas
to set the render options.
Change the Render Mode
to Screen Space - Camera
.
Then, set the Main Camera,
from the Scene tab, as the Render Camera
.
When all the render options are set, the result should look like the following image.
Select the middle
and center
option (the one that looks like a target).
Then, change the Width
to 400
, the Height
to 200
and the Pos X
to -600
.
For a proper view, navigate to the Main Camera
component and set the Y Position
to 0
, since the 3D coordinates of the hands are estimated relative to the cartesian origin point. Also, set the Z Position
to -1
to move the camera farther from the hands.
Open the webcam to get the live feed.
Check that the camera is open and available for capturing video.
To display the detected hands in 3D world space, first sort the hands by position X.
In this example, each new hand will be positioned 25 cm away from the previous one.
Close the webcam to stop the live feed, preventing further video capture.
Here is the full example code that has the same functionality as the Hand Tracking Unity plugin LightBuzz_Hand_Tracking_3D
sample.
By following these steps, you will be able to load the camera feed into your application, detect hands in real time, and finally, render these detections in 3D world space.
For this example, you'll need the , and prefabs.
Drag and drop the and prefabs into the Canvas.
Drag and drop the prefab on the Hierarchy
pane (but not into the Canvas).
To see the 3D detections, the needs to be outside of the Canvas.
For this example, we'll need a to get the frames, an to draw camera texture and a to visualize the detected hands.
To display the hand detections in the 3D world space we need to adjust the .
Navigate to the prefab, under the Canvas
, go to the Aspect Ratio Fitter
section and change Aspect Mode
to None
.
Then, go to the Rect Transform
section of the prefab and click on the blue cross to open the Anchor Presets
.
The suggested Rect Transform
settings are designed to reduce the size of the and reposition it to the side. Feel free to experiment and set the settings to different values.
Finally, return to the MonoBehaviour
script and instantiate a to detect hands.
Load the new frame from the _webcam
object onto the to show the live feed to your users.
Pass the Texture2D
object from the camera frame to the for processing.
The will analyze the texture and detect any hands present in the image.
By default, the positions the root joint of any detected hand at the origin point (0, 0, 0) within the 3D world space. To track multiple hands and visualize them clearly within the 3D environment, you need to provide an offset value in meters for each hand. Provided with an offset list, the will reposition each hand according to the offset value.
Then, simply pass the hand detections and the offsets to the . It will manage the rendering and updates required to accurately depict the hands in 3D world space based on the detection data provided.
Dispose of the object to ensure that all associated resources are released.