I just had a thought to use kinect as a animation joint recorder for characters, instead of those messy un realistic key points. Is this possible, has anybody had any success with this?
As much as i know it is possible to animate most of the body using Kinect except the fingers…
There are several applications out there on the internet which enable the kinect to be used to record BVH files, mostly for Windows I believe. The BVH can then be handled like any other BVH motion capture data in your 3d software, e.g. rigged up to a model and exported as a FBX which can then be used in Unity. Data from Kinect will not be as comprehensive as traditional motion capture systems, but its good enough for some stuff and the price setup ease are hard to beat.
This afternoon I am looking at whether alternative approaches are possible, e.g. capture the data directly inside Unity.
An example of BVH software for Kinect:
Success with mission to capture the animation data inside Unity rather than go the BVH route.
I used the Physics to Animation Tool from the Asset Store, which is discussed in this thread:
http://forum.unity3d.com/threads/85766-Physics-To-Animation-Tool-Released-!
For my particular setup Im not using the OpenNI Unity wrapper to get the Kinect data to Unity, Im using a separate app that sends OpenNI Nite skeletal tracking joint positions as OSC messages which a script in Unity can read. These points control the position of some game objects which I have connected to a rag doll model using spring joints. I use Physics To Animation to turn the ragdoll joint positions rotations into an animation clip. I can then use this clip to control animation of a non-ragdoll version of the model (remove unity joints rigid bodies, leaving just the skinned mesh and transform joints). I am doing it this way for other reasons, Im sure the same method would also work fine with the proper OpenNI Unity wrapper and non-ragdoll rigged model.
Out of interest, has anyone managed to achieve the same effect via webcam?
v0.49 of brekel gives good results with only one cam ( kinect ) but don’t go in too angled direction ( don’t try to do a 360° rotation and don’t have shoulders and hips aligned in kinect view direction or you’ll have uggly interpolations ). you still have to reduce keys and replace your root. if you use ipisoft ( 400 to 1000 dollar ) you can use up to 6 cameras and capture precisely each moves without rotations limitations ( all joints are visible using multiple cameras )
People have experimented with such things for many years. Results for skeletal tracking have tended to be very far from satisfactory under a wide range of conditions, unless additional technology is employed, such as having markers attached to certain body parts. The depth data which the Kinect provides is a real difference maker.
If the users environmental conditions are ok, then webcams can be used to do fairly basic tracking, such as head tracking, or working out overall area in a 2d space that your body limbs occupies, kinda ok. Over the last decade or so there have been a few different attempts to use this stuff in a gaming context, with results that can be fun to play with for a while, but the Kinect similar devices are a major step forward.