Revision history [back]

cv::StereoBM (in rtabmap) may not be able to generate a disparity image, so the point cloud sent to PCLVisualizer is empty and not added, causing other functions assuming that the cloud is there to throw this error. If you send your stereo images to stereo_img_proc, can it produce a valid disparity image and a point cloud? Can you also update your question with an example of left and right images (those sent to rtabmap) with corresponding camera_info?

cv::StereoBM (in rtabmap) may not be able to generate a disparity image, so the point cloud sent to PCLVisualizer is empty and not added, causing other functions assuming that the cloud is there to throw this error. If you send your stereo images to stereo_img_proc, can it produce a valid disparity image and a point cloud? Can you also update your question with an example of left and right images (those sent to rtabmap) with corresponding camera_info?

cv::StereoBM (in rtabmap) may not be able to generate a disparity image, so the point cloud sent to PCLVisualizer is empty and not added, causing other functions assuming that the cloud is there to throw this error. If you send your stereo images to stereo_img_proc, can it produce a valid disparity image and a point cloud? Can you also update your question with an example of left and right images (those sent to rtabmap) with corresponding camera_info?

EDIT

After trying the bag, here some problems:

1. Baseline is large: from right camera_info above, -(-263.97/212.49) = 1.24 m. What is the actual baseline between cameras (measured manually or from spec)?
2. Because of (1), the scale of the point cloud generated is huge. By default on older rtabmap versions, the point clouds are filtered up to 4 meters in 3d Rendering settings. Set max depth to 0 to avoid filtering the clouds.
3. The images (when comparing left and right) are not correctly rectified, there are some pixel errors along Y axis in images.
4. You may need to set frame_id of images/camera_infos to "left_optical_frame" for example, and add a static_transform_publisher base_link -> left_optical_frame with transform "0 0 0 -1.570796327 0 -1.570796327" to that point clouds are correctly transformed in ROS coordinates.

A screenshot of the problems: Note the y error on left with superposed left and right images. On right the grid is 1 meter, frustums of both cameras are shown for reference (looking upward as there is no optical rotation).