lots of nan data when convert pointcloud2 to pcl::PointXYZ

edit

HI all, I use Ubuntu 14.04 LTI with ROS Indigo. Then I simulate kinect in Gazebo:

 <xacro:macro name="kinect_camera" params="name parent *origin">
  <xacro:kinect_camera_model name="${name}" parent="${parent}">
    <xacro:insert_block name="origin" />
  </xacro:kinect_camera_model>

  <gazebo reference="${name}_depth_frame">
    <sensor type="depth" name="${name}">
      <update_rate>20</update_rate>
      <camera>
        <horizontal_fov>${60 * M_PI/180.0}</horizontal_fov>
        <image>
          <format>R8G8B8</format>
          <width>640</width>
          <height>480</height>
        </image>
        <clip>
          <near>0.05</near>
          <far>3</far>
        </clip>
      </camera>
      <plugin name="${name}_camera_controller" filename="libgazebo_ros_openni_kinect.so">
        <alwaysOn>true</alwaysOn>
        <updateRate>20</updateRate>
        <imageTopicName>${name}/rgb/image_raw</imageTopicName>
        <cameraInfoTopicName>${name}/rgb/camera_info</cameraInfoTopicName>
        <depthImageTopicName>${name}/depth/image_raw</depthImageTopicName>
        <depthImageCameraInfoTopicName>${name}/depth/camera_info</depthImageCameraInfoTopicName>
        <pointCloudTopicName>${name}/depth/points</pointCloudTopicName>
        <pointCloudCutoff>0.5</pointCloudCutoff>
        <frameName>${name}_depth_optical_frame</frameName>
        <distortion_k1>0.0</distortion_k1>
        <distortion_k2>0.0</distortion_k2>
        <distortion_k3>0.0</distortion_k3>
        <distortion_t1>0.0</distortion_t1>
        <distortion_t2>0.0</distortion_t2>
      </plugin>
    </sensor>
  </gazebo>
</xacro:macro>

Then I convert pointcloud2 to pcl::PointXYZ follow the link:pcl/overview

3.1 For a subscriber that receives pcl::PointCloud<t> objects directly:

My code:

void callback(const pcl::PointCloud<pcl::PointXYZ>::ConstPtr& position)  
{  
     const char *path="/home/lin/Desktop/distance.dat";  
     ofstream file(path);  
     for(int i=0; i<307200;i++)
 {    
      cout << "(x,y,z) = " << position -> points[i] << std::endl;  
      file << position -> points[i];  
  }  
}  


int main(int argc, char** argv){

    ros::init(argc, argv, "distance_publisher");
    ros::NodeHandle nh;

    std::string topic = nh.resolveName("camera/depth/points");
    uint32_t queue_size = 1;

    // create a templated subscriber
    ros::Subscriber sub = nh.subscribe<pcl::PointCloud<pcl::PointXYZ> > (topic, queue_size, callback);

        ros::Rate r(0.1);
        while(nh.ok()){

          ros::spinOnce(); 
          r.sleep();

        }

}

`
I get lots of NAN data , I just want to know when convert pointcloud2 to pcl::PointXYZ, it will remove nan data automatically or I need to filter it.