Detection of dynamic objects in the SLAM algorithm

Hello!

I would like to solve the detection of dynamic objects in the map during SLAM. I use the lidarSLAM() object to create the map. I have not read anywhere that this algorithm is used in the addScan(slamObj, scans {i}); function directly used. Could you recommend me how to solve it or direct me? By dynamic object I mean, for example, a human moving around a mapping robot during mapping

Thanks

[Sandip Kumar]

During performing SLAM, ones aim is to build a map of the environment. Dynamic objects should be filtered out from the SLAM map.

The algorithm implemented in lidarSLAM(...) ensures that unless you see a particular location as occupied for multiple steps of addScan, it will not be be registered to be a occupied cell in the map. So in case of dynamic objects, as they keep moving, they do not usually interfere with the actual map creation problem. Sometimes if the robot (which is generating the scans) is moving faster than the dynamic object, then there might be false occupied cells in the map.

Let us know if you need further clarification.

[Michal Mihálik]

Thanks for the explanation

Now I will clarify a bit the situation I would like to solve. The robot walks through the space it maps (in this case it is a corridor) and in addition to the robot, a person passes or still follows it. this results in it being rendered when viewing the SLAM object (added using addscan ()). after the grid map is subsequently generated, these dynamic objects appear as inaccessible zones. you can see this in the picture in the appendix. Specifically, in this case, I am behind the robot during its mapping and I am marked by a yellow highlighter in the picture. My idea is to identify such a dynamic object dynamically, highlight it on the map and then delete it when generating the map. How would you advise me to get started? or what part should I focus on? graphs and pictures are in the appendix

Answer 1

[sparsh garg]

Hey michael you can look into DynaSlam,there is a lot of research work being done in this field

https://bertabescos.github.io/DynaSLAM/

[Michal Mihálik]

Thanks for advice

However, this article focuses more on camera and 3D mapping, while I focus on 2D and lidar.

[sparsh garg]

you can look at iros workshops I am not sure of the correct name,but there are some examples which focus on this