SLAM with a camera without odometry

Question

I am new to the computing world. I recently bought a Kinect v2 for windows for making a self-driving car to navigate in the indoor areas. I have a small RC car with simple functionalities of start, stop, right, left, backward but without any sensors mounted on the car. The car doesn't provide any data about the number of rotations or other odometry related data. Q1. Is there any particular SLAM that can be helpful to solve the above problem given the constraints of camera being Kinect v2 and a car without odometry?

Since, I intend to use the car in the indoor areas therefore the map of the environment is quite fixed (constant). I plan to create a map of the environment using kinect v2 and then later on compare the real time movement against that map in order to navigate. I found Hector SLAM as an alternate to gmapping but it uses laser scanner. Q2. Can I fake the kinect v2 camera as a laser scanner for Hector SLAM?

Q3. Kinect v2 requires AC power which stops the car from being a true autonomous driving RC car. Is there any way around that?

All suggestions are welcome :D

Note - Kinect v2, Hector SLAM, gmapping, turtle bot

---

Originally posted by pallavbakshi on ROS Answers with karma: 39 on 2017-01-13

Post score: 0

Answer 1

Question 1

There is a library called RGBDslam which can use the Kinect, and doesn't require odometry. This has been tested on Fuerte and Indigo, and the github repository also has a Kinetic branch.

For the ros wiki page: http://wiki.ros.org/rgbdslam

For the github repo: https://github.com/felixendres/rgbdslam_v2

Question 2

I assume if the above works you don't need to fake the laser scanner?

Question 3

According to a forum, the power supply is rated for 12V, 2.67A. 12V batteries are not uncommon, but you'll have to be a bit more careful about the 2.67A rating. You may also need to solder your own connectors.

---

Originally posted by petern3 with karma: 186 on 2017-01-16

This answer was ACCEPTED on the original site

Post score: 1

---

Original comments

Comment by pallavbakshi on 2017-01-17:
Hi @petern3,

thank you so much for your answer. I am trying to use RGBDSlam as well. Will update the results if it works. Do you have any idea about how can I create a 2D map using RGBDSlam and can we use that map with the ROS navigation stack?

Comment by petern3 on 2017-01-18:
I've had a quick look around, and found http://wiki.ros.org/depthimage_to_laserscan, which should be able to be used with something like gmapping. There's also this tutorial, although it may be outdated.

Comment by petern3 on 2017-01-18:
If you're just doing this, then you won't need to use RGBDslam. It's a good option, as 2D path-planning is much easier.

Comment by pallavbakshi on 2017-01-21:
I think I will be going with gmapping. I had another question about gmapping with Kinect v2 though. I was confused about the drivers. Do I need iai_kinect2 as well? I have already installed libfreenect2 though.

Comment by petern3 on 2017-01-22:
I haven't used the Kinect myself, so I don't know for sure. If you can't get the depth stream from the camera you could try installing both, otherwise I think openni2 will do the trick.