How to merge overlapping bounding boxes into one?

Question

amir 2016-3-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/270864-how-to-merge-overlapping-bounding-boxes-into-one

回答： Anand 2016-3-3

I have used the "FrontalFaceCart" and "ProfileFace" detectors in parallel and in some cases I got overlapping boxes. How can I merge them into one? And any idea why it's not detecting the second person in the attached photo?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Anand 2016-3-3

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/270864-how-to-merge-overlapping-bounding-boxes-into-one#answer_212156

You could do something similar to what the MergeThreshold option in vision.CascadeObjectDetector does. The algorithm used is described in brief here:

Documentation for vision.CascadeObjectDetector

Look under Algorithms for Merge Detection Threshold.

Essentially, you could do something like this:

If two bounding boxes overlap significantly, merge them by computing the merged bounding box at the mid point of the two detections, with size as the average of the sizes of the individual detections. You can use bboxOverlapRatio to compute the overlap ratio.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

Walter Roberson 2016-3-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/270864-how-to-merge-overlapping-bounding-boxes-into-one#answer_211961

Add the box width minus 1 to get the final box index. Now take the minimum of the left coordinates and the maximum of the right coordinates to do a horizontal merge. Likewise minimum of bottom and maximum of top to do a vertical merge. You can convert those back to widths and heights if you want: subtract left from right and add 1 to get the width.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

amir 2016-3-3

I don't get how your method works, would you please explain it more? One more thing is how can I make sure that there are overlapping boxes in each time of my detection?

请先登录，再进行评论。