Inverse sensor model stereo-vision

[Matthias Domnik]

Hello everyone,

to the best of my knowlege, this problem hasn't been discussed here, yet. Maybe it isn't a problem and I just leak of knowlege in this case. However, here we go:

Foreword:

In the paper "OctoMap: An Efficient Probabilistic 3D Mapping Framework Based on Octrees" an inverse sensor model for laser scanners has been presented. So, there are two values one for a hit and one for a miss,respectively. For my understanding this assumes, that the measurement error is constant over the whole area. I'm not an expert for laser scanners, so maybe this might be a good approximation. 

I would like to use a stereo setup with the ROS implementation octomap_server. With a stereo setup the meaurement error rises quadrtically with the the distance. At least the two values won't be constant rather a function of the measured distance (for a simple case).

According to one of the creators (Armin Hornung) stereo vision is suited for tthe usage with OctoMap (https://answers.ros.org/question/45963/does-octomap-library-suit-for-stereo-vision/). Now finally my question: Is it just an approximation to speed up the execution time (simple model --> good performance) or did I have to reimplement the octomap_server for my own to provide a function for the hit/miss probabilities?

I'm looking forward to responses. :-)

[Armin Hornung]

Hello Matthias,

Am 11.05.19 um 10:45 schrieb 'Matthias Domnik' via OctoMap developers
and users discussion:

> Now finally my question: Is it just an approximation to speed up the
> execution time (simple model --> good performance) or did I have to
> reimplement the octomap_server for my own to provide a function for
> the hit/miss probabilities?

Yes to both questions! You can either use the simple model as
approximation or implement your own sensor model.

Best regards,
Armin

[Matthias Domnik]

Hello Armin,

please excuse my (very) late response. However, your answer is pretty clear. Thank you for this. 

For now I will go with the fast solution but I hope I will find some time in the future to create a more acurate inverse model for stereo vision.

Thanks again and best regards,

Matthias