This MR enlarges the crop rect.
More than one code in the crop rect is resolved by taking the first code whose middle is inside the detected head-contour.
NOTE: I just took the value of 2 @t.schroedter has written here when he started. I don't know if we would want to further discuss that value or try out different value to find an ideal mix between: everything gets detected and we don't waste computational power. Also, it might be interesting to check if working on the whole image and then matching the codes to people like done here, when multiple codes are detected, is more efficient.
This MR of course changes the test data.
Closes #172 (closed)