c++ - 使用 OpenCV 和 C++ 进行全身检测

Question

我在我的 AR.Drone 上使用 opencv 进行全身检测，想知道 minSize 和 maxSize 参数的良好设置/限制是什么？到目前为止，我将它们留空，我的代码似乎可以很好地检测到人。我还将 minNeighbors 参数设置为 7。每次运行我的代码时，我都会得到大约 20% 的误报。我希望设置大小参数将减少误报的数量。除尺寸参数外，对其他参数设置的任何输入或建议都会有所帮助

Answer 1

The size parameters, in this case, are scene dependent. You have to check the results on several images/sequences and see the distribution of sizes of both true positives and false positives. After observing the range of values for each category, you can adjust the minSize and maxSize.

Most likely, you will come to a point where you still have some false positives and some false negatives. At this point, you can decide the threshold values by considering which is more costly to have; false positives, or false negatives.

Another approach is to post-process the results returned by the recogniser, and remove false positives based on some other criteria.