Invisible is a real-time markerless motion capture (mocap) software by Move AI, freeing talent from wearing suits or equipment, opening up solutions to common virtual set problems. Traditionally, motion capture has been the preserve of high budget films, but Invisible aims to make mocap accessible to a much broader audience including XR and low to mid budget VP and live events.

Deep learning models have been used to teach the system how to detect key points on the human body within 2D images from multiple cameras, and then reconstruct the motion in 3D. This data is then run through a local, real-time neural network to apply biomechanical and kinematic models to ensure a lifelike representation of the actor/s and their movement is achieved. This data can then be streamed into a 3D engine to puppeteer a virtual avatar in real-time, or simply provide effects such as life-like shadows for a performer on stage.

The Invisible product is provided as software only, please see the below hardware requirements necessary to run the software.

Depending on the size of the space you are capturing in, we support a minimum amount of cameras, as well as lens type to ensure stable tracking throughout the space. The supported capture volumes have been categorised into Small, Medium and Large and the details of these sizes, as well as the minimum/maximum hardware requirements are outlined below.

The camera layout measurements are based on the camera positions themselves, surrounding the space you want to capture within.

Every setup differs - The optimal focal length for your setup will be based on both the distance of the camera from the capture volume and the size of the volume. The greater the focal length, the more zoomed in the image will be, which will be more suitable when the cameras are farther away from the volume. In general, it is best to position the cameras further back from the stage (without obstruction), so that you can avoid using wide angle lenses, as they often experience large amounts of distortion in the image. If you'd like more information about which lens model we recommend for a specific volume size, check out the Volume Configurator.

The Invisible software comes with three preset intrinsic lens calibrations (2.8mm, 3.5mm & 4mm). For optimal configuration it is recommended that you capture your own intrinsics using the Invisible software, as the intrinsics may vary for each camera & lens pairing.

Please note: The current supported camera requires a CS-mount lens. To use a C-mount lens with these cameras, you will need a CS to C-mount adapter.

A USB dongle will be provided with the Software, assigned with a licence configured based on the Software sale terms. The Invisible software requires the dongle to be plugged in at all times in order to operate.

SpinView is the software provided by the camera manufacturer that allows you to manage the camera settings and check the connection on the network.

If the camera names are red, the IPs are not on the same subnet - right click and select ‘Force IP to resolve’ to correct this.

Using the below image as a guide, orientate the cameras to give the best possible view of capture volume, with the least wasted space in the frame of each camera. To achieve the best MoCap, ensure you can see your actors fully head-to-toe in as many locations as possible.

The Invisible system requires two sets of calibration data in order to operate the Mocap mode. The intrinsic calibration tells the system about the camera matrix and distortion coefficients, so that it can correctly interpret and un-distort the image (the image shown in the GUI will be the raw, distorted image, until Mocap is activated). This will only need to be done once, as long as the same camera & lens are paired together in the future and the configuration of the lens has not changed.

The extrinsic calibration tells the system where the cameras are positioned and how they are orientated, in order to combine the 2D tracking from each image and triangulate the actor(s) within the volume. This will need to be done every time a camera moves and can only be captured once the intrinsics have been provided.

When the mocap mode is activated, you can use the ‘view’ toggles to change the view modes, such as 2D/3D view, camera previews on/off and 3D overlay on/off.

When in the 2D view, you can see the 3D data overlaid on each camera's preview, or turn off the camera previews to see the 3D overlay solely from the camera's perspective.

When in the 3D view, you can see the 3D representation of the actor and the cameras within the environment. To navigate, use the WASD keys to translation, and press and hold the left mouse click to rotate.

The Software comes with a Live Link plugin for Unreal Engine, so that you can stream the mocap data and map it to your characters in real-time.

Note: The origin location of the Invisible System will be defined by the start location of the actor during calibration. This will then need to be aligned with the origin in Unreal.

During mocap, you can click ‘Start recording’ on each actor track in the Actor Tracks list to begin recording the fbx of their motion. When you’re done, hit export and you can find the fbx files within the project directory.

​

All logs will be saved and dated in the Project folder. For any support requests, include the logs folder with the enquiry. Reach out to [email protected]