Week 5-7 Realsense D435 and SLAM using RGBD sensors
1. ROS
Proceed to the installation of ROS-kinetics. Since I have Ubuntu 18 and ROS-Kinetics is only available for Ubuntu 16, I had to reinstall everything on Ubuntu 16.
Learn about ROS platform, filesystem (rosbag), ROS Computation Graph Level (similar to microservices architecture), and start testing the ROS executables examples from that comes with SD-SLAM.
Start working with the realsense D435 RGBD sensor.The objective is to create a simple program that reads information from the realsense and shows it on a window (PyQt + ROS)
After installing the Intel Realsense Drivers for the D435 and the ROS wrapper, I find some problems with the installation of the drivers using the realsense-viewer app that allows using the intel realsense drivers to visualize the streams from the Realsense D435. The RGB stream can be perfectly seen but when the Depth stream is show the application freezes and stop working. I have to found out a solution.
2. RGBD research state of the art
Search information related to the SLAM problem using RGBD sensors. There is not too much documents related to SLAM using only depth sensor information, but a lot about RGBD (visual SLAM, using feature extraction on the RGB image and then using the depth information for those features in order to improve the visual SLAM algorithms).
But I found some articles that seems interesting and talk about the SLAM from point of view of depth information:
- Depth Camera Based Indoor Mobile Robot Localization and Navigation (RANSAC on the point cloud to find planes on the 3D environment)
- 3D pose estimation and mapping with time-of-flight cameras (Uses RGBD but also has information and problems of the TOF sensors)
- Direct Depth SLAM (the same approach that we are trying to make here)
After reading Direct Depth SLAM, Ive found a lot of techniques such as range flow constrain equation, ICP and more that Ill try to understand and read more about to be able to implement them and evaluate their performance.