How is Matching Scored?

[HID_Carl]

Hello Good Day!
I would like to ask about how matching is scored? I am working on a pick and place project at present and due to the small gap of the tool and the work jig during picking, i have to improve the alignment of the matching result. One of the method that i was able to improve the alignment was to reduce the target object to a simpler shape. But with that method reduced the matching score and now i have to explain to the boss why it is more accurate yet lower in score.

the original target object model is a 4 pipes extruded from an “I” shaped bar (in cyan lower left).

I remodeled it so that it will only have the middle 2 pipes and a section of the bar, filled the hole in the pipe and added chamfer to represent the shape of weld bead of the actual work.

The mis-alignment is exaggerated and represented by the simple shapes below the zoom-in of point clouds. The scores on the first model (mis-aligned) averages to 95% while the 2nd model (somewhat aligned) averages to 85%. How did this happen? or can you recommend an article that i should read to enlighten my self?

Your answer will be much appreciated! Thank You!

[Bad_Dragon]

The top point cloud of pipes will always be matched no matter which model you use (original or remodeled) , and the side point cloud of pipes will never – camera just can’t capture that part.

Matching is scored by the proportion of the matched part relative to the complete model.
If there are 100 points in the reshape model, the unmatched part is about 15 point, so you get the score about 0.85.
And there are 1000 points in the orignal model, the unmatched part is also 15 point, so you get the score about 0.985.

So if the proportion of unmatched part is larger, the score will be lower.

And why is the score of mis-aligned model so high? Maybe you can adjust the confidence setting, change the result verification degree to Ultra-high, it will reduce the score of mis-aligned and keep the score for aligned.