robot_pose_ekf wrong output with "good input"

edit

I'm not be able to understand why I get a wrong pose from robot_pose_ekf. I'm using a kinect on a table and connected to the workstation to find the turtlebot position orientation and publish the result in /vo. Kinect is tracking a pattern marker and find a pose. I want to use odometry and imu to have a better result and have a pose when kinect can't see the marker.

My issue is /vo, /odom, and imu/data has similar orientation but ekf find something very different and I can't understand why.

Here is my tf tree to see how I work:

/vo:

header:    seq: 479   stamp: 
    secs: 1370526479
    nsecs: 1198623   frame_id: odom child_frame_id: base_footprint pose:   pose: 
    position: 
      x: 0.109848357611
      y: 0.0184851177529
      z: 0.00108856573807
    orientation: 
      x: -0.0155068317083
      y: 0.00236792210117
      z: -0.726064768505
      w: 0.687447367477   covariance: [0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 10000.1, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 10000.1, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1e-06]

/odom

header: 
  seq: 5144
  stamp: 
    secs: 1370526478
    nsecs: 985132932
  frame_id: odom
child_frame_id: base_footprint
pose: 
  pose: 
    position: 
      x: 0.122778090562
      y: 0.0180749352772
      z: 0.0
    orientation: 
      x: 0.0
      y: 0.0
      z: -0.754709580223
      w: 0.656059028991
  covariance: [0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1000.0]

/imu/data:

header: 
  seq: 5154
  stamp: 
    secs: 1370526479
    nsecs: 318391084
  frame_id: gyro_link
orientation: 
  x: 0.0
  y: 0.0
  z: -0.641532116852
  w: 0.767096175878
orientation_covariance: [1000000.0, 0.0, 0.0, 0.0, 1000000.0, 0.0, 0.0, 0.0, 1e-06]

and robot_pose_ekf/odom:

header: 
  seq: 3340
  stamp: 
    secs: 1370526479
    nsecs: 618319034
  frame_id: odom
pose: 
  pose: 
    position: 
      x: 0.0979321075308
      y: -0.0259189421082
      z: -0.00496891294622
    orientation: 
      x: 3.68249858201e-06
      y: -4.82461122176e-05
      z: 0.970055184809
      w: -0.242884614753
  covariance: [5.493486503712613e-07, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 5.493486503712613e-07, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0987278074026108e-06, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 10.877878440724317, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 10.877878440724317, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 5.493333772345698e-10]

And how I construct the odometry msg:

 void readAndSendVo()
  {
    tf::TransformListener lisVo;
    tf::StampedTransform transformVo;
    nav_msgs::Odometry vo;
    std ...