- Fixed to work with arbitrary size images, not only 640x480.
- Regression tests for multiple image sizes.
- Fixed cosmetic problem in update of small images.
New package image_rotate, for visualizing camera streams in an orientation that is more intuitive to the user than the hardware-constrained orientation of the physical camera. This is particularly helpful, for example, to show images from the PR2's forearm cameras with a consistent up direction, despite the fact that the forearms need to rotate in arbitrary ways during manipulation.
- Removed WG block matcher option. OpenCV's implementation is always used.
Mono calibration now reports linear error of the rectified checkerboard corners (<<Ticket(ros-pkg 4035)>>).
- Updated cameracheck to work with stricter topic name checking (#4036).
This release includes bug-fixes from 1.0.1 and 1.0.2. Other changes listed below.
- camera_hammer utility for testing a set_camera_info service call.
- Workaround for exiting when the user closes the window on Linux systems (#3026).
- Source files and executables now go in src/ and bin/ (#3978).
- Force min_disparity to be zero when using WG block matcher, which doesn't support non-zero values.
- Explicitly zero out border pixels that tend to have junk when using OpenCV block matcher. (Probably fixed in more recent revisions of OpenCV)
- Made OpenCV the default block matcher, preparing to remove WG reference implementation.
Specific error message on receiving image with ambiguous 8UC3 encoding.
Can initialize calibrators from CameraInfo messages (#3891).
- Changed default stereo_image_proc settings to give good stereo results on PR2 data.
- calibration no longer makes buggy assumptions about corner spacing always being .108m
- Added a simple calibration script (camera_calibrate_from_disk.py) that loads a set of images from disk and performs monocular calibration returning the camera intrinsics.
- Set projection matrix P correctly in monocular calibration.
- Warn if color topic requested but raw image data is grayscale.
- Allow using either the OpenCV block matcher or the Willow Garage reference implementation. Should give roughly the same results with vision_opencv 1.1 or greater.
Added points2 topic with the new sensor_msgs/PointCloud2 format.
- Upped texture_threshold maximum to 10000; high values needed for simulated stereo images.