segmentlidardata segmentation issue (computer vision toolbox)

I tested this function (segmentlidardata) to segment data from a lidar dataset taken in marine environment. Input used are 1m for distance threshold and 5° for angle threshold (default value). I noticed that some portion of point cluster are well segmented, other unfortunately are over-segmented. In addition I tried also the matlab example of segmentlidardata and I noticed that the same issue is notable. For istance I noticed that a over-segmentation of a road sign is present (but not for all road signs). Could be a little bug in the matlab implementation?

For better understanding of my issue I attached the screenshots of over-segmentations.

This image is refered to the matlab example, you can see the over-segmentation of the road sign (central red circle) and the well segmentation on the right lower side.

You can see also an over-segmentation problem regarding the street on the left side. I don't understand why this over-segmentation happens beacause for example the street is continue and respect the distance and angle threshold. In addition such a fact happens only in some portion of the poitn cloud.

The over-segmentation issue is very clearly notable in my point cloud where you can observe in the lower part the correct segmentation, contrarly in the upper part you can see the problem of a huge over-segmentation, also for the 2 pedestrian. What can i do regarding this issue? Am i doing it wrong?