CC-RGBD-SLAM -- Centralized Collaborative RGB-D SLAM

CC-RGBD-SLAM is a modified verison of CCM-SLAM, and it supports RGB-D mode.

Modifiction

The modifiction mainly includes three aspects.

1. Add the RGB-D mode in the depth branch.
2. Use efficient image transmission to solve the problem of limited bandwidth.
3. Add a priori map to the display module of rviz, so that we can clearly observe the clients' position on the map.

RGB-D mode

• The branch depth replaced the monocular mode before to RGB-D mode, and the sensor needs to be equipped with a depth camera. The client receives the color image and depth image from the camera, and the tracking module basically follows the RGB-D mode of ORB-SLAM2.
• It fundamentally solves the problem of scale uncertainty of the monocular camera, allowing us to locate on an actual map.

Efficient image transmission

Former CCM-SLAM client receives the original image of the camera sensor, which wil cause excessive communication bandwidth resources and cause the image to freeze during real-time operation.

• Therefore, due to the limited bandwidth, efficient image transmission is particularyly important. We use the image_transport package of ROS to compress the image topic.
• Our camera sensor is Intel RealSense D435i, and we set the frame rate to 30fps and the image size to $640\times480$. Then We check the original image topic "/camera/color/image_raw" and the compressed image topic "/camera/color/image_raw/compressed" in ROS, and see that the bandwidth is 28MB/s and 60KB/s respectively.
• Therefore we use compressed image as communication topics for transmission, and the occupied bandwidth is greatly compressed.

Display module with priori map

...