I've met several issues which need to be resolved before finishing the move, @asmorkalov , @opencv-alalek , @vpisarev , please take a look.

A) Dependency on opencv_dnn

Currently objdetect optionally depends on opencv_dnn, so that algorithms using DNN will print warnings when OpenCV has not been built with DNN support. Barcode module also can optionally use neural network for super-resolution (same as in wechat_qrcode module), but it also uses dnn::NMSBoxes function for detection directly. I see two options here:

1. move NMSBoxes to the opencv_core imgproc
2. make objdetect dependency on dnn obligatory

B) Inconvenient module API

Barcode class returns results into a three vectors which have to be iterated together afterwards:

bool detectAndDecode(
    InputArray img,
    CV_OUT std::vector<std::string> &decoded_info, // decoded strings - NUM items
    CV_OUT std::vector<BarcodeType> &decoded_type, // code type (e.g. EAN-13), basically a vector of ints - NUM items
    OutputArray points = noArray()) const; // corners - 4*NUM items

I think it will be better to refactor this interface to look sililar to wechat_qrcode, so that it will be possible to unify all interfaces later:

std::vector<std::string> detectAndDecode(InputArray img, OutputArrayOfArrays points = noArray());

Note: code type can be encoded in the first one or two symbols of returned string or returned in a separate vector or tied to each string using a pair or packed in a special structure

Note: I took a look at the #23264 and it also has slight inconsistency between single and multi interfaces:

std::string detectAndDecode(
    InputArray img,
    OutputArray points=noArray(),
    OutputArray straight_qrcode = noArray());
// but
bool detectAndDecodeMulti(
    InputArray img, 
    CV_OUT std::vector<std::string>& decoded_info,
    OutputArray points = noArray(),
    OutputArrayOfArrays straight_qrcode = noArray()) const;

C) Algorithm separation

Currently there are two algorithms used internally in the barcode implementation which can be moved outside:

• special binarization can be moved to other binarization methods (imgproc, cv::threshold, cv::adaptiveThreshold)
• DNN-based super-resolution is copied from wechat_qrcode and can be moved to the dnn module, similar to FaceDetectorYN.

Created #23711 which moves NMSBoxes from dnn to imgproc, thus we will be able to move barcode detector to objdetect while keeping its OPTIONAL dependency on dnn.