I want to discuss some problem and improvement idea on using iPi Studio with more than one camera. I am using two cameras right now. I think there are many difficulties to use this software that can be solved easily if there are more features on it.

First, it is very difficult to make a room without reflective things on it. The typical problem that is very hard to solve is the reflectivity of the floor, which always shown every time the light marker touching near the ground. We can tackle it with video postprocessing software like After Effects but it will time consuming and not practical. I think the best solution for this is by creating more than one Region of Interest to calibrate, so that we can jump from block to block to hide every bad frame which contain more than one marker. If we can set many Region of Interest, we don't have to mask unwanted light and render again every scene that we have made.

Second, it is very difficult to scaling the appearance with only one parameter for Actor Tab (his Height). The underlying assumption that ideal man should have proportional body will make big distortion when the scene being tracked. I give one example. My hand is longer than the hand of the Actor at the height of my height (1.70 m), so that the Actor's hand lying on the appearance of my arms (not my hands). It makes big error when iPi Studio tracking my arms! There many people out there which has non proportional body. We can import target character from external animation software to solve this, but the best solution is to give length parameters for every bone. With multiple bone length parameters, we can adjust the Actor exactly equal with the Appearance. This solution will improve the tracking of the arms movement, which is main weakness of iPi Studio.

Third, on Camera Section, there is one parameter which is cannot be set precisely. The Field of View parameter will be better if it can be filled with directly typing the number and not by moving the slider. Many camera already have their number, like my logitech camera, but it cannot be set exactly the same number with the slider. The inaccuracy of the slider sometimes more than 0.05 degree.

Fourth, I have triple core processor, but only one processor doing their job (not 100%) when calibrating or tracking process. Is there any possibilities to speed up the process with multithreading?

Thank you very much for your answer and suggestion.

iwansuryo

Thanks for your feedback.
Here are developers comments:

1) Up to 30% incorrect marker detections should not represent a problem. Software cross-checks marker detections using information from all cameras and ignores incorrect detections. So far as you have a bright marker and it gets correctly detected in about 70% of frames, everything should work fine and you should not care about tricks like multiple regions of interest.

We will release more demos and tutorials soon to illustrate this.

2) We are working on more sophisticated actor body shape models. Current model is a reasonable default that should work for most humans. It is true that there is some variation in human body shape, but things like height to hands wingspan ratio remain pretty constant (like 2-3% variation for different people).

http://en.wikipedia.org/wiki/Vitruvian_Man

If you observe a large disproportion when you overlay actor model and video, this is most probably caused by incorrect multi-camera calibration. You can verify actor body proportions by shooting a single-camera video of actor in T-pose and overlaying it with actor model using iPi Studio.

3) The latest version is heavily multi-threaded so it should make use of all 3 cores of yoru CPU. Please keep in mind that our software does most of calculations on video card. So you need a fast video card for running it. You mentioned you have GeForce 9500 - it's an entry-level video card and it is relatively slow. Consider using at least something like GeForce 9800 GT or Radeon 4850 (both cards should cost around $100) - those are dramatically faster cards.