2019-12-18 23:08:07 +00:00
|
|
|
package gocv
|
|
|
|
|
|
|
|
/*
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include "features2d.h"
|
|
|
|
*/
|
|
|
|
import "C"
|
|
|
|
import (
|
|
|
|
"image/color"
|
|
|
|
"reflect"
|
|
|
|
"unsafe"
|
|
|
|
)
|
|
|
|
|
|
|
|
// AKAZE is a wrapper around the cv::AKAZE algorithm.
|
|
|
|
type AKAZE struct {
|
|
|
|
// C.AKAZE
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewAKAZE returns a new AKAZE algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d8/d30/classcv_1_1AKAZE.html
|
|
|
|
//
|
|
|
|
func NewAKAZE() AKAZE {
|
|
|
|
return AKAZE{p: unsafe.Pointer(C.AKAZE_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close AKAZE.
|
|
|
|
func (a *AKAZE) Close() error {
|
|
|
|
C.AKAZE_Close((C.AKAZE)(a.p))
|
|
|
|
a.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using AKAZE.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (a *AKAZE) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.AKAZE_Detect((C.AKAZE)(a.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// DetectAndCompute keypoints and compute in an image using AKAZE.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#a8be0d1c20b08eb867184b8d74c15a677
|
|
|
|
//
|
|
|
|
func (a *AKAZE) DetectAndCompute(src Mat, mask Mat) ([]KeyPoint, Mat) {
|
|
|
|
desc := NewMat()
|
|
|
|
ret := C.AKAZE_DetectAndCompute((C.AKAZE)(a.p), src.p, mask.p, desc.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret), desc
|
|
|
|
}
|
|
|
|
|
|
|
|
// AgastFeatureDetector is a wrapper around the cv::AgastFeatureDetector.
|
|
|
|
type AgastFeatureDetector struct {
|
|
|
|
// C.AgastFeatureDetector
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewAgastFeatureDetector returns a new AgastFeatureDetector algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d7/d19/classcv_1_1AgastFeatureDetector.html
|
|
|
|
//
|
|
|
|
func NewAgastFeatureDetector() AgastFeatureDetector {
|
|
|
|
return AgastFeatureDetector{p: unsafe.Pointer(C.AgastFeatureDetector_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close AgastFeatureDetector.
|
|
|
|
func (a *AgastFeatureDetector) Close() error {
|
|
|
|
C.AgastFeatureDetector_Close((C.AgastFeatureDetector)(a.p))
|
|
|
|
a.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using AgastFeatureDetector.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (a *AgastFeatureDetector) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.AgastFeatureDetector_Detect((C.AgastFeatureDetector)(a.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BRISK is a wrapper around the cv::BRISK algorithm.
|
|
|
|
type BRISK struct {
|
|
|
|
// C.BRISK
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewBRISK returns a new BRISK algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d8/d30/classcv_1_1AKAZE.html
|
|
|
|
//
|
|
|
|
func NewBRISK() BRISK {
|
|
|
|
return BRISK{p: unsafe.Pointer(C.BRISK_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close BRISK.
|
|
|
|
func (b *BRISK) Close() error {
|
|
|
|
C.BRISK_Close((C.BRISK)(b.p))
|
|
|
|
b.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using BRISK.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (b *BRISK) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.BRISK_Detect((C.BRISK)(b.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// DetectAndCompute keypoints and compute in an image using BRISK.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#a8be0d1c20b08eb867184b8d74c15a677
|
|
|
|
//
|
|
|
|
func (b *BRISK) DetectAndCompute(src Mat, mask Mat) ([]KeyPoint, Mat) {
|
|
|
|
desc := NewMat()
|
|
|
|
ret := C.BRISK_DetectAndCompute((C.BRISK)(b.p), src.p, mask.p, desc.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret), desc
|
|
|
|
}
|
|
|
|
|
|
|
|
// FastFeatureDetectorType defines the detector type
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/df/d74/classcv_1_1FastFeatureDetector.html#a4654f6fb0aa4b8e9123b223bfa0e2a08
|
|
|
|
type FastFeatureDetectorType int
|
|
|
|
|
|
|
|
const (
|
|
|
|
//FastFeatureDetectorType58 is an alias of FastFeatureDetector::TYPE_5_8
|
|
|
|
FastFeatureDetectorType58 FastFeatureDetectorType = 0
|
|
|
|
//FastFeatureDetectorType712 is an alias of FastFeatureDetector::TYPE_7_12
|
2020-09-06 12:53:53 +00:00
|
|
|
FastFeatureDetectorType712 FastFeatureDetectorType = 1
|
2019-12-18 23:08:07 +00:00
|
|
|
//FastFeatureDetectorType916 is an alias of FastFeatureDetector::TYPE_9_16
|
2020-09-06 12:53:53 +00:00
|
|
|
FastFeatureDetectorType916 FastFeatureDetectorType = 2
|
2019-12-18 23:08:07 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
// FastFeatureDetector is a wrapper around the cv::FastFeatureDetector.
|
|
|
|
type FastFeatureDetector struct {
|
|
|
|
// C.FastFeatureDetector
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewFastFeatureDetector returns a new FastFeatureDetector algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/df/d74/classcv_1_1FastFeatureDetector.html
|
|
|
|
//
|
|
|
|
func NewFastFeatureDetector() FastFeatureDetector {
|
|
|
|
return FastFeatureDetector{p: unsafe.Pointer(C.FastFeatureDetector_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewFastFeatureDetectorWithParams returns a new FastFeatureDetector algorithm with parameters
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/df/d74/classcv_1_1FastFeatureDetector.html#ab986f2ff8f8778aab1707e2642bc7f8e
|
|
|
|
//
|
|
|
|
func NewFastFeatureDetectorWithParams(threshold int, nonmaxSuppression bool, typ FastFeatureDetectorType) FastFeatureDetector {
|
|
|
|
return FastFeatureDetector{p: unsafe.Pointer(C.FastFeatureDetector_CreateWithParams(C.int(threshold), C.bool(nonmaxSuppression), C.int(typ)))}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close FastFeatureDetector.
|
|
|
|
func (f *FastFeatureDetector) Close() error {
|
|
|
|
C.FastFeatureDetector_Close((C.FastFeatureDetector)(f.p))
|
|
|
|
f.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using FastFeatureDetector.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (f *FastFeatureDetector) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.FastFeatureDetector_Detect((C.FastFeatureDetector)(f.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GFTTDetector is a wrapper around the cv::GFTTDetector algorithm.
|
|
|
|
type GFTTDetector struct {
|
|
|
|
// C.GFTTDetector
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewGFTTDetector returns a new GFTTDetector algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/df/d21/classcv_1_1GFTTDetector.html
|
|
|
|
//
|
|
|
|
func NewGFTTDetector() GFTTDetector {
|
|
|
|
return GFTTDetector{p: unsafe.Pointer(C.GFTTDetector_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close GFTTDetector.
|
|
|
|
func (a *GFTTDetector) Close() error {
|
|
|
|
C.GFTTDetector_Close((C.GFTTDetector)(a.p))
|
|
|
|
a.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using GFTTDetector.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (a *GFTTDetector) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.GFTTDetector_Detect((C.GFTTDetector)(a.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// KAZE is a wrapper around the cv::KAZE algorithm.
|
|
|
|
type KAZE struct {
|
|
|
|
// C.KAZE
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewKAZE returns a new KAZE algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d3/d61/classcv_1_1KAZE.html
|
|
|
|
//
|
|
|
|
func NewKAZE() KAZE {
|
|
|
|
return KAZE{p: unsafe.Pointer(C.KAZE_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close KAZE.
|
|
|
|
func (a *KAZE) Close() error {
|
|
|
|
C.KAZE_Close((C.KAZE)(a.p))
|
|
|
|
a.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using KAZE.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (a *KAZE) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.KAZE_Detect((C.KAZE)(a.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// DetectAndCompute keypoints and compute in an image using KAZE.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#a8be0d1c20b08eb867184b8d74c15a677
|
|
|
|
//
|
|
|
|
func (a *KAZE) DetectAndCompute(src Mat, mask Mat) ([]KeyPoint, Mat) {
|
|
|
|
desc := NewMat()
|
|
|
|
ret := C.KAZE_DetectAndCompute((C.KAZE)(a.p), src.p, mask.p, desc.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret), desc
|
|
|
|
}
|
|
|
|
|
|
|
|
// MSER is a wrapper around the cv::MSER algorithm.
|
|
|
|
type MSER struct {
|
|
|
|
// C.MSER
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewMSER returns a new MSER algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d3/d28/classcv_1_1MSER.html
|
|
|
|
//
|
|
|
|
func NewMSER() MSER {
|
|
|
|
return MSER{p: unsafe.Pointer(C.MSER_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close MSER.
|
|
|
|
func (a *MSER) Close() error {
|
|
|
|
C.MSER_Close((C.MSER)(a.p))
|
|
|
|
a.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using MSER.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (a *MSER) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.MSER_Detect((C.MSER)(a.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// ORB is a wrapper around the cv::ORB.
|
|
|
|
type ORB struct {
|
|
|
|
// C.ORB
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewORB returns a new ORB algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d7/d19/classcv_1_1AgastFeatureDetector.html
|
|
|
|
//
|
|
|
|
func NewORB() ORB {
|
|
|
|
return ORB{p: unsafe.Pointer(C.ORB_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close ORB.
|
|
|
|
func (o *ORB) Close() error {
|
|
|
|
C.ORB_Close((C.ORB)(o.p))
|
|
|
|
o.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using ORB.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (o *ORB) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.ORB_Detect((C.ORB)(o.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// DetectAndCompute detects keypoints and computes from an image using ORB.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#a8be0d1c20b08eb867184b8d74c15a677
|
|
|
|
//
|
|
|
|
func (o *ORB) DetectAndCompute(src Mat, mask Mat) ([]KeyPoint, Mat) {
|
|
|
|
desc := NewMat()
|
|
|
|
ret := C.ORB_DetectAndCompute((C.ORB)(o.p), src.p, mask.p, desc.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret), desc
|
|
|
|
}
|
|
|
|
|
|
|
|
// SimpleBlobDetector is a wrapper around the cv::SimpleBlobDetector.
|
|
|
|
type SimpleBlobDetector struct {
|
|
|
|
// C.SimpleBlobDetector
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// SimpleBlobDetector_Params is a wrapper around the cv::SimpleBlobdetector::Params
|
|
|
|
type SimpleBlobDetectorParams struct {
|
|
|
|
p C.SimpleBlobDetectorParams
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewSimpleBlobDetector returns a new SimpleBlobDetector algorithm
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d7a/classcv_1_1SimpleBlobDetector.html
|
|
|
|
//
|
|
|
|
func NewSimpleBlobDetector() SimpleBlobDetector {
|
|
|
|
return SimpleBlobDetector{p: unsafe.Pointer(C.SimpleBlobDetector_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewSimpleBlobDetectorWithParams returns a new SimpleBlobDetector with custom parameters
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d7a/classcv_1_1SimpleBlobDetector.html
|
|
|
|
//
|
|
|
|
func NewSimpleBlobDetectorWithParams(params SimpleBlobDetectorParams) SimpleBlobDetector {
|
|
|
|
return SimpleBlobDetector{p: unsafe.Pointer(C.SimpleBlobDetector_Create_WithParams(params.p))}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close SimpleBlobDetector.
|
|
|
|
func (b *SimpleBlobDetector) Close() error {
|
|
|
|
C.SimpleBlobDetector_Close((C.SimpleBlobDetector)(b.p))
|
|
|
|
b.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewSimpleBlobDetectorParams returns the default parameters for the SimpleBobDetector
|
|
|
|
func NewSimpleBlobDetectorParams() SimpleBlobDetectorParams {
|
|
|
|
return SimpleBlobDetectorParams{p: C.SimpleBlobDetectorParams_Create()}
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetBlobColor sets the blobColor field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetBlobColor(blobColor int) {
|
|
|
|
p.p.blobColor = C.uchar(blobColor)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetBlobColor gets the blobColor field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetBlobColor() int {
|
|
|
|
return int(p.p.blobColor)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetFilterByArea sets the filterByArea field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetFilterByArea(filterByArea bool) {
|
|
|
|
p.p.filterByArea = C.bool(filterByArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetFilterByArea gets the filterByArea field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetFilterByArea() bool {
|
|
|
|
return bool(p.p.filterByArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetFilterByCircularity sets the filterByCircularity field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetFilterByCircularity(filterByCircularity bool) {
|
|
|
|
p.p.filterByCircularity = C.bool(filterByCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetFilterByCircularity gets the filterByCircularity field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetFilterByCircularity() bool {
|
|
|
|
return bool(p.p.filterByCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetFilterByColor sets the filterByColor field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetFilterByColor(filterByColor bool) {
|
|
|
|
p.p.filterByColor = C.bool(filterByColor)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetFilterByColor gets the filterByColor field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetFilterByColor() bool {
|
|
|
|
return bool(p.p.filterByColor)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetFilterByConvexity sets the filterByConvexity field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetFilterByConvexity(filterByConvexity bool) {
|
|
|
|
p.p.filterByConvexity = C.bool(filterByConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetFilterByConvexity gets the filterByConvexity field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetFilterByConvexity() bool {
|
|
|
|
return bool(p.p.filterByConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetFilterByInertia sets the filterByInertia field
|
|
|
|
func (p *SimpleBlobDetectorParams) SetFilterByInertia(filterByInertia bool) {
|
|
|
|
p.p.filterByInertia = C.bool(filterByInertia)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetFilterByInertia gets the filterByInertia field
|
|
|
|
func (p *SimpleBlobDetectorParams) GetFilterByInertia() bool {
|
|
|
|
return bool(p.p.filterByInertia)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMaxArea sets the maxArea parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMaxArea(maxArea float64) {
|
|
|
|
p.p.maxArea = C.float(maxArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMaxArea sets the maxArea parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMaxArea() float64 {
|
|
|
|
return float64(p.p.maxArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMaxCircularity sets the maxCircularity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMaxCircularity(maxCircularity float64) {
|
|
|
|
p.p.maxCircularity = C.float(maxCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMaxCircularity sets the maxCircularity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMaxCircularity() float64 {
|
|
|
|
return float64(p.p.maxCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMaxConvexity sets the maxConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMaxConvexity(maxConvexity float64) {
|
|
|
|
p.p.maxConvexity = C.float(maxConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMaxConvexity sets the maxConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMaxConvexity() float64 {
|
|
|
|
return float64(p.p.maxConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMaxInertiaRatio sets the maxInertiaRatio parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMaxInertiaRatio(maxInertiaRatio float64) {
|
|
|
|
p.p.maxInertiaRatio = C.float(maxInertiaRatio)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMaxInertiaRatio sets the maxCConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMaxInertiaRatio() float64 {
|
|
|
|
return float64(p.p.maxInertiaRatio)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMaxThreshold sets the maxThreshold parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMaxThreshold(maxThreshold float64) {
|
|
|
|
p.p.maxThreshold = C.float(maxThreshold)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMaxThreshold sets the maxCConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMaxThreshold() float64 {
|
|
|
|
return float64(p.p.maxThreshold)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinArea sets the minArea parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinArea(minArea float64) {
|
|
|
|
p.p.minArea = C.float(minArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinArea sets theinArea parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinArea() float64 {
|
|
|
|
return float64(p.p.minArea)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinCircularity sets the minCircularity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinCircularity(minCircularity float64) {
|
|
|
|
p.p.minCircularity = C.float(minCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinCircularity sets the minCircularity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinCircularity() float64 {
|
|
|
|
return float64(p.p.minCircularity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinConvexity sets the minConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinConvexity(minConvexity float64) {
|
|
|
|
p.p.minConvexity = C.float(minConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinConvexity sets the minConvexity parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinConvexity() float64 {
|
|
|
|
return float64(p.p.minConvexity)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinDistBetweenBlobs sets the minDistBetweenBlobs parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinDistBetweenBlobs(minDistBetweenBlobs float64) {
|
|
|
|
p.p.minDistBetweenBlobs = C.float(minDistBetweenBlobs)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinDistBetweenBlobs sets the minDistBetweenBlobs parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinDistBetweenBlobs() float64 {
|
|
|
|
return float64(p.p.minDistBetweenBlobs)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinInertiaRatio sets the minInertiaRatio parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinInertiaRatio(minInertiaRatio float64) {
|
|
|
|
p.p.minInertiaRatio = C.float(minInertiaRatio)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinInertiaRatio sets the minInertiaRatio parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinInertiaRatio() float64 {
|
|
|
|
return float64(p.p.minInertiaRatio)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinRepeatability sets the minRepeatability parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinRepeatability(minRepeatability int) {
|
|
|
|
p.p.minRepeatability = C.size_t(minRepeatability)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinInertiaRatio sets the minRepeatability parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinRepeatability() int {
|
|
|
|
return int(p.p.minRepeatability)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinThreshold sets the minThreshold parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetMinThreshold(minThreshold float64) {
|
|
|
|
p.p.minThreshold = C.float(minThreshold)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinThreshold sets the minInertiaRatio parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetMinThreshold() float64 {
|
|
|
|
return float64(p.p.minThreshold)
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetMinThreshold sets the minThreshold parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) SetThresholdStep(thresholdStep float64) {
|
|
|
|
p.p.thresholdStep = C.float(thresholdStep)
|
|
|
|
}
|
|
|
|
|
|
|
|
// GetMinThreshold sets the minInertiaRatio parameter for SimpleBlobDetector_Params
|
|
|
|
func (p *SimpleBlobDetectorParams) GetThresholdStep() float64 {
|
|
|
|
return float64(p.p.thresholdStep)
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using SimpleBlobDetector.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (b *SimpleBlobDetector) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.SimpleBlobDetector_Detect((C.SimpleBlobDetector)(b.p), src.p)
|
|
|
|
defer C.KeyPoints_Close(ret)
|
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// getKeyPoints returns a slice of KeyPoint given a pointer to a C.KeyPoints
|
|
|
|
func getKeyPoints(ret C.KeyPoints) []KeyPoint {
|
|
|
|
cArray := ret.keypoints
|
|
|
|
length := int(ret.length)
|
|
|
|
hdr := reflect.SliceHeader{
|
|
|
|
Data: uintptr(unsafe.Pointer(cArray)),
|
|
|
|
Len: length,
|
|
|
|
Cap: length,
|
|
|
|
}
|
|
|
|
s := *(*[]C.KeyPoint)(unsafe.Pointer(&hdr))
|
|
|
|
|
|
|
|
keys := make([]KeyPoint, length)
|
|
|
|
for i, r := range s {
|
|
|
|
keys[i] = KeyPoint{float64(r.x), float64(r.y), float64(r.size), float64(r.angle), float64(r.response),
|
|
|
|
int(r.octave), int(r.classID)}
|
|
|
|
}
|
|
|
|
return keys
|
|
|
|
}
|
|
|
|
|
|
|
|
// BFMatcher is a wrapper around the the cv::BFMatcher algorithm
|
|
|
|
type BFMatcher struct {
|
|
|
|
// C.BFMatcher
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewBFMatcher returns a new BFMatcher
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d3/da1/classcv_1_1BFMatcher.html#abe0bb11749b30d97f60d6ade665617bd
|
|
|
|
//
|
|
|
|
func NewBFMatcher() BFMatcher {
|
|
|
|
return BFMatcher{p: unsafe.Pointer(C.BFMatcher_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewBFMatcherWithParams creates a new BFMatchers but allows setting parameters
|
|
|
|
// to values other than just the defaults.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d3/da1/classcv_1_1BFMatcher.html#abe0bb11749b30d97f60d6ade665617bd
|
|
|
|
//
|
|
|
|
func NewBFMatcherWithParams(normType NormType, crossCheck bool) BFMatcher {
|
|
|
|
return BFMatcher{p: unsafe.Pointer(C.BFMatcher_CreateWithParams(C.int(normType), C.bool(crossCheck)))}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close BFMatcher
|
|
|
|
func (b *BFMatcher) Close() error {
|
|
|
|
C.BFMatcher_Close((C.BFMatcher)(b.p))
|
|
|
|
b.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// KnnMatch Finds the k best matches for each descriptor from a query set.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/db/d39/classcv_1_1DescriptorMatcher.html#aa880f9353cdf185ccf3013e08210483a
|
|
|
|
//
|
|
|
|
func (b *BFMatcher) KnnMatch(query, train Mat, k int) [][]DMatch {
|
|
|
|
ret := C.BFMatcher_KnnMatch((C.BFMatcher)(b.p), query.p, train.p, C.int(k))
|
|
|
|
defer C.MultiDMatches_Close(ret)
|
|
|
|
|
|
|
|
return getMultiDMatches(ret)
|
|
|
|
}
|
|
|
|
|
2021-01-17 18:07:37 +00:00
|
|
|
// FlannBasedMatcher is a wrapper around the the cv::FlannBasedMatcher algorithm
|
|
|
|
type FlannBasedMatcher struct {
|
|
|
|
// C.FlannBasedMatcher
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewFlannBasedMatcher returns a new FlannBasedMatcher
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/dc/de2/classcv_1_1FlannBasedMatcher.html#ab9114a6471e364ad221f89068ca21382
|
|
|
|
//
|
|
|
|
func NewFlannBasedMatcher() FlannBasedMatcher {
|
|
|
|
return FlannBasedMatcher{p: unsafe.Pointer(C.FlannBasedMatcher_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close FlannBasedMatcher
|
|
|
|
func (f *FlannBasedMatcher) Close() error {
|
|
|
|
C.FlannBasedMatcher_Close((C.FlannBasedMatcher)(f.p))
|
|
|
|
f.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// KnnMatch Finds the k best matches for each descriptor from a query set.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/db/d39/classcv_1_1DescriptorMatcher.html#aa880f9353cdf185ccf3013e08210483a
|
|
|
|
//
|
|
|
|
func (f *FlannBasedMatcher) KnnMatch(query, train Mat, k int) [][]DMatch {
|
|
|
|
ret := C.FlannBasedMatcher_KnnMatch((C.FlannBasedMatcher)(f.p), query.p, train.p, C.int(k))
|
|
|
|
defer C.MultiDMatches_Close(ret)
|
|
|
|
|
|
|
|
return getMultiDMatches(ret)
|
|
|
|
}
|
|
|
|
|
2019-12-18 23:08:07 +00:00
|
|
|
func getMultiDMatches(ret C.MultiDMatches) [][]DMatch {
|
|
|
|
cArray := ret.dmatches
|
|
|
|
length := int(ret.length)
|
|
|
|
hdr := reflect.SliceHeader{
|
|
|
|
Data: uintptr(unsafe.Pointer(cArray)),
|
|
|
|
Len: length,
|
|
|
|
Cap: length,
|
|
|
|
}
|
|
|
|
s := *(*[]C.DMatches)(unsafe.Pointer(&hdr))
|
|
|
|
|
|
|
|
keys := make([][]DMatch, length)
|
|
|
|
for i := range s {
|
|
|
|
keys[i] = getDMatches(C.MultiDMatches_get(ret, C.int(i)))
|
|
|
|
}
|
|
|
|
return keys
|
|
|
|
}
|
|
|
|
|
|
|
|
func getDMatches(ret C.DMatches) []DMatch {
|
|
|
|
cArray := ret.dmatches
|
|
|
|
length := int(ret.length)
|
|
|
|
hdr := reflect.SliceHeader{
|
|
|
|
Data: uintptr(unsafe.Pointer(cArray)),
|
|
|
|
Len: length,
|
|
|
|
Cap: length,
|
|
|
|
}
|
|
|
|
s := *(*[]C.DMatch)(unsafe.Pointer(&hdr))
|
|
|
|
|
|
|
|
keys := make([]DMatch, length)
|
|
|
|
for i, r := range s {
|
|
|
|
keys[i] = DMatch{int(r.queryIdx), int(r.trainIdx), int(r.imgIdx),
|
|
|
|
float64(r.distance)}
|
|
|
|
}
|
|
|
|
return keys
|
|
|
|
}
|
|
|
|
|
|
|
|
// DrawMatchesFlag are the flags setting drawing feature
|
|
|
|
//
|
|
|
|
// For further details please see:
|
|
|
|
// https://docs.opencv.org/master/de/d30/structcv_1_1DrawMatchesFlags.html
|
|
|
|
type DrawMatchesFlag int
|
|
|
|
|
|
|
|
const (
|
|
|
|
// DrawDefault creates new image and for each keypoint only the center point will be drawn
|
|
|
|
DrawDefault DrawMatchesFlag = 0
|
|
|
|
// DrawOverOutImg draws matches on existing content of image
|
2020-09-06 12:53:53 +00:00
|
|
|
DrawOverOutImg DrawMatchesFlag = 1
|
2019-12-18 23:08:07 +00:00
|
|
|
// NotDrawSinglePoints will not draw single points
|
2020-09-06 12:53:53 +00:00
|
|
|
NotDrawSinglePoints DrawMatchesFlag = 2
|
2019-12-18 23:08:07 +00:00
|
|
|
// DrawRichKeyPoints draws the circle around each keypoint with keypoint size and orientation
|
2020-09-06 12:53:53 +00:00
|
|
|
DrawRichKeyPoints DrawMatchesFlag = 3
|
2019-12-18 23:08:07 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
// DrawKeyPoints draws keypoints
|
|
|
|
//
|
|
|
|
// For further details please see:
|
|
|
|
// https://docs.opencv.org/master/d4/d5d/group__features2d__draw.html#gab958f8900dd10f14316521c149a60433
|
|
|
|
func DrawKeyPoints(src Mat, keyPoints []KeyPoint, dst *Mat, color color.RGBA, flag DrawMatchesFlag) {
|
|
|
|
cKeyPointArray := make([]C.struct_KeyPoint, len(keyPoints))
|
|
|
|
|
|
|
|
for i, kp := range keyPoints {
|
|
|
|
cKeyPointArray[i].x = C.double(kp.X)
|
|
|
|
cKeyPointArray[i].y = C.double(kp.Y)
|
|
|
|
cKeyPointArray[i].size = C.double(kp.Size)
|
|
|
|
cKeyPointArray[i].angle = C.double(kp.Angle)
|
|
|
|
cKeyPointArray[i].response = C.double(kp.Response)
|
|
|
|
cKeyPointArray[i].octave = C.int(kp.Octave)
|
|
|
|
cKeyPointArray[i].classID = C.int(kp.ClassID)
|
|
|
|
}
|
|
|
|
|
|
|
|
cKeyPoints := C.struct_KeyPoints{
|
|
|
|
keypoints: (*C.struct_KeyPoint)(&cKeyPointArray[0]),
|
|
|
|
length: (C.int)(len(keyPoints)),
|
|
|
|
}
|
|
|
|
|
|
|
|
scalar := C.struct_Scalar{
|
2020-09-06 12:53:53 +00:00
|
|
|
val1: C.double(color.B),
|
2019-12-18 23:08:07 +00:00
|
|
|
val2: C.double(color.G),
|
2020-09-06 12:53:53 +00:00
|
|
|
val3: C.double(color.R),
|
2019-12-18 23:08:07 +00:00
|
|
|
val4: C.double(color.A),
|
|
|
|
}
|
|
|
|
|
|
|
|
C.DrawKeyPoints(src.p, cKeyPoints, dst.p, scalar, C.int(flag))
|
|
|
|
}
|
2020-09-06 12:53:53 +00:00
|
|
|
|
|
|
|
// SIFT is a wrapper around the cv::SIFT algorithm.
|
|
|
|
// Due to the patent having expired, this is now in the main OpenCV code modules.
|
|
|
|
type SIFT struct {
|
|
|
|
// C.SIFT
|
|
|
|
p unsafe.Pointer
|
|
|
|
}
|
|
|
|
|
|
|
|
// NewSIFT returns a new SIFT algorithm.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d5/d3c/classcv_1_1xfeatures2d_1_1SIFT.html
|
|
|
|
//
|
|
|
|
func NewSIFT() SIFT {
|
|
|
|
return SIFT{p: unsafe.Pointer(C.SIFT_Create())}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Close SIFT.
|
|
|
|
func (d *SIFT) Close() error {
|
|
|
|
C.SIFT_Close((C.SIFT)(d.p))
|
|
|
|
d.p = nil
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Detect keypoints in an image using SIFT.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#aa4e9a7082ec61ebc108806704fbd7887
|
|
|
|
//
|
|
|
|
func (d *SIFT) Detect(src Mat) []KeyPoint {
|
|
|
|
ret := C.SIFT_Detect((C.SIFT)(d.p), C.Mat(src.Ptr()))
|
2021-01-17 18:07:37 +00:00
|
|
|
defer C.KeyPoints_Close(ret)
|
2020-09-06 12:53:53 +00:00
|
|
|
|
|
|
|
return getKeyPoints(ret)
|
|
|
|
}
|
|
|
|
|
|
|
|
// DetectAndCompute detects and computes keypoints in an image using SIFT.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d0/d13/classcv_1_1Feature2D.html#a8be0d1c20b08eb867184b8d74c15a677
|
|
|
|
//
|
|
|
|
func (d *SIFT) DetectAndCompute(src Mat, mask Mat) ([]KeyPoint, Mat) {
|
|
|
|
desc := NewMat()
|
|
|
|
ret := C.SIFT_DetectAndCompute((C.SIFT)(d.p), C.Mat(src.Ptr()), C.Mat(mask.Ptr()),
|
|
|
|
C.Mat(desc.Ptr()))
|
2021-01-17 18:07:37 +00:00
|
|
|
defer C.KeyPoints_Close(ret)
|
2020-09-06 12:53:53 +00:00
|
|
|
|
|
|
|
return getKeyPoints(ret), desc
|
|
|
|
}
|
2021-01-17 18:07:37 +00:00
|
|
|
|
|
|
|
// DrawMatches draws matches on combined train and querry images.
|
|
|
|
//
|
|
|
|
// For further details, please see:
|
|
|
|
// https://docs.opencv.org/master/d4/d5d/group__features2d__draw.html#gad8f463ccaf0dc6f61083abd8717c261a
|
|
|
|
func DrawMatches(img1 Mat, kp1 []KeyPoint, img2 Mat, kp2 []KeyPoint, matches1to2 []DMatch, outImg *Mat, matchColor color.RGBA, singlePointColor color.RGBA, matchesMask []byte, flags DrawMatchesFlag) {
|
|
|
|
kp1arr := make([]C.struct_KeyPoint, len(kp1))
|
|
|
|
kp2arr := make([]C.struct_KeyPoint, len(kp2))
|
|
|
|
|
|
|
|
for i, kp := range kp1 {
|
|
|
|
kp1arr[i].x = C.double(kp.X)
|
|
|
|
kp1arr[i].y = C.double(kp.Y)
|
|
|
|
kp1arr[i].size = C.double(kp.Size)
|
|
|
|
kp1arr[i].angle = C.double(kp.Angle)
|
|
|
|
kp1arr[i].response = C.double(kp.Response)
|
|
|
|
kp1arr[i].octave = C.int(kp.Octave)
|
|
|
|
kp1arr[i].classID = C.int(kp.ClassID)
|
|
|
|
}
|
|
|
|
|
|
|
|
for i, kp := range kp2 {
|
|
|
|
kp2arr[i].x = C.double(kp.X)
|
|
|
|
kp2arr[i].y = C.double(kp.Y)
|
|
|
|
kp2arr[i].size = C.double(kp.Size)
|
|
|
|
kp2arr[i].angle = C.double(kp.Angle)
|
|
|
|
kp2arr[i].response = C.double(kp.Response)
|
|
|
|
kp2arr[i].octave = C.int(kp.Octave)
|
|
|
|
kp2arr[i].classID = C.int(kp.ClassID)
|
|
|
|
}
|
|
|
|
|
|
|
|
cKeyPoints1 := C.struct_KeyPoints{
|
|
|
|
keypoints: (*C.struct_KeyPoint)(&kp1arr[0]),
|
|
|
|
length: (C.int)(len(kp1)),
|
|
|
|
}
|
|
|
|
|
|
|
|
cKeyPoints2 := C.struct_KeyPoints{
|
|
|
|
keypoints: (*C.struct_KeyPoint)(&kp2arr[0]),
|
|
|
|
length: (C.int)(len(kp2)),
|
|
|
|
}
|
|
|
|
|
|
|
|
dMatchArr := make([]C.struct_DMatch, len(matches1to2))
|
|
|
|
|
|
|
|
for i, dm := range matches1to2 {
|
|
|
|
dMatchArr[i].queryIdx = C.int(dm.QueryIdx)
|
|
|
|
dMatchArr[i].trainIdx = C.int(dm.TrainIdx)
|
|
|
|
dMatchArr[i].imgIdx = C.int(dm.ImgIdx)
|
|
|
|
dMatchArr[i].distance = C.float(dm.Distance)
|
|
|
|
}
|
|
|
|
|
|
|
|
cDMatches := C.struct_DMatches{
|
|
|
|
dmatches: (*C.struct_DMatch)(&dMatchArr[0]),
|
|
|
|
length: (C.int)(len(matches1to2)),
|
|
|
|
}
|
|
|
|
|
|
|
|
scalarMatchColor := C.struct_Scalar{
|
|
|
|
val1: C.double(matchColor.R),
|
|
|
|
val2: C.double(matchColor.G),
|
|
|
|
val3: C.double(matchColor.B),
|
|
|
|
val4: C.double(matchColor.A),
|
|
|
|
}
|
|
|
|
|
|
|
|
scalarPointColor := C.struct_Scalar{
|
|
|
|
val1: C.double(singlePointColor.B),
|
|
|
|
val2: C.double(singlePointColor.G),
|
|
|
|
val3: C.double(singlePointColor.R),
|
|
|
|
val4: C.double(singlePointColor.A),
|
|
|
|
}
|
|
|
|
|
|
|
|
mask := make([]C.char, len(matchesMask))
|
|
|
|
|
|
|
|
cByteArray := C.struct_ByteArray{
|
|
|
|
length: (C.int)(len(matchesMask)),
|
|
|
|
}
|
|
|
|
|
|
|
|
if len(matchesMask) > 0 {
|
|
|
|
cByteArray = C.struct_ByteArray{
|
|
|
|
data: (*C.char)(&mask[0]),
|
|
|
|
length: (C.int)(len(matchesMask)),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
C.DrawMatches(img1.p, cKeyPoints1, img2.p, cKeyPoints2, cDMatches, outImg.p, scalarMatchColor, scalarPointColor, cByteArray, C.int(flags))
|
|
|
|
}
|