Making sure detected plane is gound plane

[bit-pirate]

The RANSAC approach to detect the ground plane is straight-forward and usually works well. However, with the current implementation the ground plane is assumed to be always the largest plane in the point cloud. This is not always true - imagine standing in front of a wall or large flat obstacle (e.g. table). Hence, we need to make sure that the selected plane is actually the ground plane.

[kentsommer]

Few ideas:

• Discard the RANSAC approach and do something along the lines of a calibration step before the Robot will be used. Once a mathematical model of the floor plane is created any point that does not fit the model can be thrown out. This would keep walls from possibly messing up the algorithm, however, this does not handle ramps well.
• Use a mixture of RANSAC and the above approach. Do a calibration step, determine a mathmatical model for the ground plane and then if at any point the RANSAC determined plane is not within the bounds for the modeled plane, it is safe to assume it is a ramp/wall/etc.

[stonier]

Ramps could eventually be integrated into whatever approach we use by incorporating input from an inclinometer.

[bit-pirate]

On 11 September 2014 12:08, Daniel Stonier notifications@github.com wrote:

Ramps could eventually be integrated into whatever approach we use by incorporating input from an inclinometer.

"inclinometer"? That looks like a real aussie tool - seems to be used most commonly in 4WD and airplanes. :-D

RE topic: What about adding a parameter for now, which defines the tilt angle and use this tilt angle to determine the floor plane. Later on we add a subscriber, which allows updating the tilt parameter on runtime (e.g. using the gyro/accelerometer feedback).

— Reply to this email directly or view it on GitHub https://github.com/kentsommer/obstacle_detection/issues/5#issuecomment-55214074 .

Innovation Team Leader Yujin Robot Seoul, Republic of Korea Twitter: @yujinrobotinno Website: http://inno.yujinrobot.com Email: marcus.liebhardt@yujinrobot.com Phone: +82-70-46577073

[stonier]

Ah, this is just an expression used for a way of doing things with a 3 axis accelerometer which is typical in any AHRS system (and yes, their business uses are mostly vehicles and airplanes simply because there isn't many robots out there doing this kind of thing yet!). You just use that to work out the gravity queue and calculate the inclination from that.

I believe it can be 2 axis as well, but we'll want to know the orientation of the inclination in our case anyway.