Use find_scaled_shape_models do template matching, found online can not find even a little article about this operator, only a few find_scaled_shape_model related,
but at the same time I want to do more than one template matching, one does not want to.
Take that question yesterday, is to configure the minimum acceptable value of the score, the score value is obviously far lower than the actual score, but it is not matched on the issue.
Later, after more testing and re-read this official document found multiple templates if an acceptable minimum scores are the same, you really can write only one value, the question Where is it?
Today, the test found that the threshold is adjusted after 0.4, there should be a partial match on the match, which is very bad, this part of the score about 4.6, in general, able to match the real score on the usually above 0.8.
To solve this problem, I decided to re-adjust the threshold value of 0.5, as expected, this one does not match on the other a few should not also match the match, almost yes. . .
Problem discovery process:
to find only one halcon provided contains routines that operators found inside use only a fraction, not an array, excluded for this reason;
not the same Maxoverlap and I set, the meaning of this parameter is searchable after a maximum overlap between the template instance to set to the same value, no effect;
use halcon hdevelop matching the assistant, the same template and test image, a threshold value is adjusted to 0.5, it can be matched, which is very strange, assistant using find_scaled_shape_model, is it that these two operators themselves are not really the same?
To find, insert the code, a comparison parameters.
Ultimately found to be greedy algorithm parameters that may not be appropriate. This parameter between 0 and 1, the smaller the more careful search, took longer; the greater the less careful search, the shorter the time-consuming. Before a configuration became 0.9,
which is an official recommended that the documents generally score above 0.9, and 0.9 of the greedy algorithm configured to basically be able to find out. What a pit, I have a score Figure 0.92 there, but that is not to find out.
And with the official routines that are 0.8.
Later, through a little tune quiz, eventually determined to be 0.5, not so time-consuming because of the high demand, tentative first with this value, coupled with an acceptable minimum score of 0.5, the test for some time to see results.
Estimate this thing with a few people, why did I fancy it too ....