TOF Depth Image Encoding changed from 32FC1 to mono8

  • I have a TOF sensor.

    A little while ago, to see the depth image, I would rostopic echo /tof/depth. When I did that, the ROS message would be of the type sensor_imgs/Image and its encoding would be 32FC1.

    Today, with the latest code, I need to use rostopic echo /mpa/tof_depth to get the depth image. When I do that, the ROS message is still of the type sensor_imgs/Image but its encoding is mono8.

    Historically, if I wanted to process the depth image in ROS, I would setup a subscriber and use code like that shown below to ingest the data as a depth image. I could then access individual pixels to get the depth (in meters).

    I'm now sure how to process a depth image of encoding mono8. It seems like it would have too few bits to represent the resolution of depths possible.

    Is the mono8 encoding intentional (with the conversion to the latest mpa_to_ros)? Or is it a bug?

    If it's intentional, can you help me understand how to process this new data encoding and extract the distances of individual pixels (similar to the code posted below)?

            # get the depth image
            dep_frame = np.fromstring(, dtype=np.single)
            dep_frame = dep_frame.reshape(172, 224) 
            # Get closest value in bounding box
            object_depth_map = dep_frame[top:bottom,left:right]
            object_depth_map[object_depth_map == 0] = 9999.00
            object_depth = np.min(object_depth_map)

  • Dev Team

    MPA_to_ROS was recently updated to a new architecture, the depth image that we publish was meant to be more of a debug tool where the z value of the pointcloud is scaled from 0-255 for 0-1 meter distance, but it looks like the old mpa to ros instead published the z value of the sensor, which is floating point. To get the same old value that you were looking at, you should access the pointcloud tof message and look at the z value.

  • For anyone who runs into this same issue, I changed my subscriber to listen to /map/tof_pc (which has the PointCloud2 type). My code (to extract the z value) is this:

    # get the depth image
            dep_frame = np.fromstring(, dtype=np.single) # get numpy array
            dep_frame = dep_frame.reshape(172, 224, 3)[:,:,2]

Log in to reply