Camera_model segfault. Does it require a "real" camera_info not to crash?

edit

Hello,

I am working on a node that colors point clouds with a camera image. My strategy is to project each point on the camera image plane, and find the corresponding pixel color.

In order to do so, I use the image_geometry::PinholeCameraModel object that can initialize itself directly from a sensor_msgs::camera_info message via the function fromCameraInfo and project points via the function project3dToPixel.

The current setup I have consist of a labybug3 camera and a tilting hokuyo LRF. I haven't develop a proper driver for the camera, so for the moment, I am using the camera_1394 package using the following configuration:

<node pkg="camera1394" type="camera1394_node" name="ladybug" >
    <param name="guid" value="00b09d01009f01a2" />
    <param name="video_mode" value="1600x1200_mono8" />
    <param name="frame_id" value="ladybug0" />
    <param name="bayer_pattern" value="rggb" />
    <param name="auto_white_balance" value="2" />
</node>

As you can notice, I do not provide any calibration info, which result in uncalibrated camera_info messages. (from camera1394: If CameraInfo calibration is not available or is incompatible with the current video_mode, uncalibrated data will be provided instead.)

Which is indeed what I can observe with rostopic:

header: 
  seq: 22480
  stamp: 
    secs: 1305821789
    nsecs: 356266975
  frame_id: /ladybug0
height: 1200
width: 1600
distortion_model: ''
D: []
K: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
R: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
P: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
binning_x: 0
binning_y: 0
roi: 
  x_offset: 0
  y_offset: 0
  height: 0
  width: 0
  do_rectify: False
---

My problem is that I am able to initialize my camera model using those camera_info messages, but my program keeps crashing on the call to project3dToPixel.

Code:

cv::Point2d pixel;
pcl::PointCloud<pcl::PointXYZ> cloud_xyz;
[...]
for(int i = 0 ; i < (int) cloud_xyz.size(); i++)
{
     if (isnan (cloud_xyz.points[i].x) || isnan (cloud_xyz.points[i].y) || isnan (cloud_xyz.points[i].z))
            continue;
     ROS_INFO("point with coordinates %f, %f, %f", cloud_xyz.points[i].x, cloud_xyz.points[i].y, cloud_xyz.points[i].z);
     ROS_INFO("A");
     cv::Point3d point(cloud_xyz.points[i].x, cloud_xyz.points[i].y, cloud_xyz.points[i].z);
     ROS_INFO("B");
     pixel = cam_model.project3dToPixel(point);

Trace:

[ INFO] [1305820636.985045709]: point with coordinates 0.245843, -6.119996, 0.848742
[ INFO] [1305820636.985082376]: A
[ INFO] [1305820636.985117506]: B
Segmentation fault

I had a look inside image_geometry::PinholeCameraModel's documentation but I can't find any information on the case where the camera_info messages are 0.

My best guess so far is that the on calibrated camera_info message triggers the segfault at runtime. Does anyone has experience with this issue?

Thanks for your help

Raph