Depth sensors provide body tracking data based on embedded tracking algorithms. The accuracy of such tracking is rather low and far below requirements of professional-level motion capture solutions. Tracking provided by iPi Mocap Studio
is based on iPi Soft's algorithms and do not use tracking data from sensors. However, sensor's tracking can be effectively used to improve workflow.
We record and use tracking data from Kinect 2 sensors. The data are used for:
- Detecting number of actors
- Detecting actor heights
- Detecting T-Poses and creation of takes
- Setting initial pose for detected T-poses
- Setting inital pose for the beginnign frame of the Region-of-Interest
- Warning on presense of people in sensor view during background recording
Recording Sensor Tracking Data
iPi Recorder records tracking data provided by depth sensor for further use in iPi Mocap Studio.
As soon as embedded tracking algorithms work well only when actor faces the sensor and each sensor works indepndently from others, for T-Pose and takes creation we only use data from the first sensor.
To make the most of sensors' tracking data, please follow this recommendations during recording:
- Strike T-Pose before each take
- You have to stay in T-Pose at least for one second
- Face the first sensor in T-Pose
- If you have several actors, each of them should face the first sensor
In case when capture area does not allow to face the sensor, you can stay at an angle to sensor. Make sure that arms and legs are not occluded from sensor.
If you stand back to sensor, embedded tracking will detect gaze direction opposite to the actual, so automatically detected inital pose will be incorrect and you'll have to manually fix this in iPi Mocap Studio
Using Sensor Tracking Data
Detecting Number of Actors and Actor Heights
When creating a new project from recorded .iPiVideo file, you need to specify number of actors and their heghts. These numbers are detected based on tracking data. You just need to check them during project creation, as embedded tracking algorithms are not very accurate. Correct the numbers if necessary.
If you only have single depth sensor, you'll not have number of actors selection step, as multiple actors tracking is not supported for single depth sensor.
Actor's gender is not detected by sensor, so you'll need to set it manually for female actors.
Creating Takes and Setting Actor Poses
Takes created automatically
- On project creation takes will be created based on detected T-Poses
- Region-of-Interest will be set equal to first take
- Actor's pose will be set based on first sensor's tracking data in the beginning frame of the Region-of-Interest
- You can manually adjust takes and the Region-of-Interest
- When you change the beginning frame of the Region-of-Interest, actor's pose will be set from sensor's tracking data in the beginning frame
- You must perform Refit Pose in the beginning frame of Region-of-Interest before you start tracking
Pose can be set from sensor's tracking data only if you have not performed tracking or any manual editing of the pose in this frame. Otherwise pose will not be changed.
When you change the beginning frame of Region-of-Interest
, poses perviously set from sensor's tracking data will not be deleted. They will not affect tracking uless you perform Refit Pose
in this frame. So there's no need to manually delete them.
To get better result, you need to adjust actor's proportions before you start tracking.
If actor stands back to first sensor, his gaze direction will be set opposite from actual, and the entire pose will be incorrect. You can delete it and set manually.
If pose is not accurate enough and Refit Pose
does not fix it, you need to adjust pose manually to ensure it is close to actual depth data.