Variables that Improve SLAM

Question

I have a general question regarding SLAM as I am not an expert on the topic. What are the variables that can improve map generation?

The ones I can think of are:

1. A higher resolution camera
2. Consistent lighting conditions
3. A room with plenty of high confidence landmarks
4. A lot of processing power on the robot

Could someone help elaborate this list for me or tell me if I'm completely wrong?

Answer 1

If you are restricted to camera-only SLAM you are on a good track. You mentioned a robot, do you use its odometry?

• How good is your extrinsic robot-camera calibration?
• How good is your intrinsic camera calibration?
• Could you add a better IMU?
• Could you improve the Odometry by changing the wheels or the floor?
• Can you add other sensors like a laserscanner?
• Does your camera have a rolling shutter or global shutter?
• Do you exactly know when the image acquisition was started (e.g. because it was triggered by a hardware trigger?)
• Which opening angle does your camera have? (bigger would be better if you can handle lens distortion)
• Can you add artificial landmarks (e.g. QR-Marker) in the scene?

Answer 2

Your Question is in generall so the answers will be, too. What is your application? It is in Indoor or Outdoor? Do you have a specific SLAM in you mind to use? However, in general, I propose you do not limit your self to monocular cameras. If you are in an indoor environment I propose you RGBD cameras and in an outdoor environment, stereo cameras would be a better fit. Aso do not forget to calibrate your camera, see these two link for calibrating the camera by OpenCV and ROS. The next stepps will be odometry fusion. Using IMU, wheel encoders, LIDAR, Sonar and so on will be helpful. The robot pose EKF can help you with odometry fusion.