Obstacle Avoidance with Point Cloud Library

Question

I've been developing a system that aims at detecting obstacles (with Kinect) and then according to the Kinect perspective, finding the non-obstacle side. I am using Point Cloud Library since I am working with depth data. So far, i was able to make floor segmentation, some downsamplings (voxel grid, passthrough) and clustering (potential obstacles). Final stage of my work is to find safe route (non-obstacle side). But I am not sure how can i get the horizontal bounds (x axis) of Kinect perspective to decide the exact location of obstacles in the scene.

Any help would be appreciated

Answer 1

Ok, i found a solution. Here it is;

1-)I have divided the screen into 4 horizontal parts. Each part represents 0.75m gap that is sufficient to pass through easily by average human.

2-)To navigate user safe side, i count the number of points for each side (left or right).

3-)After 2nd step is done, i know the side that has few points. Then, in the side that has few points, i perform a sub-control. For example, if the right side has few points, then i calculate points in right side's sub sides (close right,far right).

4-) Finally, the sub-side that has less points then the other sub side is transposed to user if it satisfies the number of points criteria ( if it contains 100 or less points)

Hope it would help someone else that is working on it.