Open Source Computer Vision Library https://opencv.org/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

465 lines
20 KiB

//
// Calib3dTest.swift
//
// Created by Giles Payne on 2020/05/26.
//
import XCTest
import OpenCV
class Calib3dTest: OpenCVTestCase {
var size = Size()
override func setUp() {
super.setUp()
size = Size(width: 3, height: 3)
}
override func tearDown() {
super.tearDown()
}
func testComposeRTMatMatMatMatMatMat() throws {
let rvec1 = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try rvec1.put(row: 0, col: 0, data: [0.5302828, 0.19925919, 0.40105945] as [Float])
let tvec1 = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try tvec1.put(row: 0, col: 0, data: [0.81438506, 0.43713298, 0.2487897] as [Float])
let rvec2 = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try rvec2.put(row: 0, col: 0, data: [0.77310503, 0.76209372, 0.30779448] as [Float])
let tvec2 = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try tvec2.put(row: 0, col: 0, data: [0.70243168, 0.4784472, 0.79219002] as [Float])
let rvec3 = Mat()
let tvec3 = Mat()
let outRvec = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try outRvec.put(row: 0, col: 0, data: [1.418641, 0.88665926, 0.56020796])
let outTvec = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
try outTvec.put(row: 0, col: 0, data: [1.4560841, 1.0680628, 0.81598103])
Calib3d.composeRT(rvec1: rvec1, tvec1: tvec1, rvec2: rvec2, tvec2: tvec2, rvec3: rvec3, tvec3: tvec3)
try assertMatEqual(outRvec, rvec3, OpenCVTestCase.EPS)
try assertMatEqual(outTvec, tvec3, OpenCVTestCase.EPS)
}
func testFilterSpecklesMatDoubleIntDouble() throws {
gray_16s_1024.copy(to: dst)
let center = Point(x: gray_16s_1024.rows() / 2, y: gray_16s_1024.cols() / 2)
Imgproc.circle(img: dst, center: center, radius: 1, color: Scalar.all(4096))
try assertMatNotEqual(gray_16s_1024, dst)
Calib3d.filterSpeckles(img: dst, newVal: 1024.0, maxSpeckleSize: 100, maxDiff: 0.0)
try assertMatEqual(gray_16s_1024, dst)
}
func testFindChessboardCornersMatSizeMat() {
let patternSize = Size(width: 9, height: 6)
let corners = MatOfPoint2f()
Calib3d.findChessboardCorners(image: grayChess, patternSize: patternSize, corners: corners)
XCTAssertFalse(corners.empty())
}
func testFindChessboardCornersMatSizeMatInt() {
let patternSize = Size(width: 9, height: 6)
let corners = MatOfPoint2f()
Calib3d.findChessboardCorners(image: grayChess, patternSize: patternSize, corners: corners, flags: Calib3d.CALIB_CB_ADAPTIVE_THRESH + Calib3d.CALIB_CB_NORMALIZE_IMAGE + Calib3d.CALIB_CB_FAST_CHECK)
XCTAssertFalse(corners.empty())
}
func testFind4QuadCornerSubpix() {
let patternSize = Size(width: 9, height: 6)
let corners = MatOfPoint2f()
let region_size = Size(width: 5, height: 5)
Calib3d.findChessboardCorners(image: grayChess, patternSize: patternSize, corners: corners)
Calib3d.find4QuadCornerSubpix(img: grayChess, corners: corners, region_size: region_size)
XCTAssertFalse(corners.empty())
}
func testFindCirclesGridMatSizeMat() {
let size = 300
let img = Mat(rows:Int32(size), cols:Int32(size), type:CvType.CV_8U)
img.setTo(scalar: Scalar(255))
let centers = Mat()
XCTAssertFalse(Calib3d.findCirclesGrid(image: img, patternSize: Size(width: 5, height: 5), centers: centers))
for i in 0..<5 {
for j in 0..<5 {
let x = Int32(size * (2 * i + 1) / 10)
let y = Int32(size * (2 * j + 1) / 10)
let pt = Point(x: x, y: y)
Imgproc.circle(img: img, center: pt, radius: 10, color: Scalar(0), thickness: -1)
}
}
XCTAssert(Calib3d.findCirclesGrid(image: img, patternSize:Size(width:5, height:5), centers:centers))
XCTAssertEqual(25, centers.rows())
XCTAssertEqual(1, centers.cols())
XCTAssertEqual(CvType.CV_32FC2, centers.type())
}
func testFindCirclesGridMatSizeMatInt() {
let size:Int32 = 300
let img = Mat(rows:size, cols: size, type: CvType.CV_8U)
img.setTo(scalar: Scalar(255))
let centers = Mat()
XCTAssertFalse(Calib3d.findCirclesGrid(image: img, patternSize: Size(width: 3, height: 5), centers: centers, flags: Calib3d.CALIB_CB_CLUSTERING | Calib3d.CALIB_CB_ASYMMETRIC_GRID))
let step = size * 2 / 15
let offsetx = size / 6
let offsety = (size - 4 * step) / 2
for i:Int32 in 0...2 {
for j:Int32 in 0...4 {
let pt = Point(x: offsetx + (2 * i + j % 2) * step, y: offsety + step * j)
Imgproc.circle(img: img, center: pt, radius: 10, color: Scalar(0), thickness: -1)
}
}
XCTAssert(Calib3d.findCirclesGrid(image: img, patternSize: Size(width: 3, height: 5), centers: centers, flags: Calib3d.CALIB_CB_CLUSTERING | Calib3d.CALIB_CB_ASYMMETRIC_GRID))
XCTAssertEqual(15, centers.rows())
XCTAssertEqual(1, centers.cols())
XCTAssertEqual(CvType.CV_32FC2, centers.type())
}
func testFindHomographyListOfPointListOfPoint() throws {
let NUM:Int32 = 20
let originalPoints = MatOfPoint2f()
originalPoints.alloc(NUM)
let transformedPoints = MatOfPoint2f()
transformedPoints.alloc(NUM)
for i:Int32 in 0..<NUM {
let x:Float = Float.random(in: -50...50)
let y:Float = Float.random(in: -50...50)
try originalPoints.put(row:i, col:0, data:[x, y])
try transformedPoints.put(row:i, col:0, data:[y, x])
}
let hmg = Calib3d.findHomography(srcPoints: originalPoints, dstPoints: transformedPoints)
truth = Mat(rows: 3, cols: 3, type: CvType.CV_64F)
try truth!.put(row:0, col:0, data:[0, 1, 0, 1, 0, 0, 0, 0, 1] as [Double])
try assertMatEqual(truth!, hmg, OpenCVTestCase.EPS)
}
func testReprojectImageTo3DMatMatMat() throws {
let transformMatrix = Mat(rows: 4, cols: 4, type: CvType.CV_64F)
try transformMatrix.put(row:0, col:0, data:[0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1] as [Double])
let disparity = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_32F)
var disp = [Float].init(repeating: 0.0, count: Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize))
for i in 0..<Int(OpenCVTestCase.matSize) {
for j in 0..<Int(OpenCVTestCase.matSize) {
disp[i * Int(OpenCVTestCase.matSize) + j] = Float(i - j)
}
}
try disparity.put(row:0, col:0, data:disp)
let _3dPoints = Mat()
Calib3d.reprojectImageTo3D(disparity: disparity, _3dImage: _3dPoints, Q: transformMatrix)
XCTAssertEqual(CvType.CV_32FC3, _3dPoints.type())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.rows())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.cols())
truth = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_32FC3)
var _truth = [Float](repeating: 0.0, count: Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize * 3))
for i:Int in 0..<Int(OpenCVTestCase.matSize) {
for j:Int in 0..<Int(OpenCVTestCase.matSize) {
let start:Int = (i * Int(OpenCVTestCase.matSize) + j) * 3
_truth[start + 0] = Float(i)
_truth[start + 1] = Float(j)
_truth[start + 2] = Float(i - j)
}
}
try truth!.put(row: 0, col: 0, data: _truth)
try assertMatEqual(truth!, _3dPoints, OpenCVTestCase.EPS)
}
func testReprojectImageTo3DMatMatMatBoolean() throws {
let transformMatrix = Mat(rows: 4, cols: 4, type: CvType.CV_64F)
try transformMatrix.put(row: 0, col: 0, data: [0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1] as [Double])
let disparity = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_32F)
var disp = [Float](repeating: 0.0, count: Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize))
for i in 0..<Int(OpenCVTestCase.matSize) {
for j in 0..<Int(OpenCVTestCase.matSize) {
disp[i * Int(OpenCVTestCase.matSize) + j] = Float(i - j)
}
}
disp[0] = -.greatestFiniteMagnitude
try disparity.put(row: 0, col: 0, data: disp)
let _3dPoints = Mat()
Calib3d.reprojectImageTo3D(disparity: disparity, _3dImage: _3dPoints, Q: transformMatrix, handleMissingValues: true)
XCTAssertEqual(CvType.CV_32FC3, _3dPoints.type())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.rows())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.cols())
truth = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_32FC3)
var _truth = [Float](repeating: 0.0, count:Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize * 3))
for i in 0..<Int(OpenCVTestCase.matSize) {
for j in 0..<Int(OpenCVTestCase.matSize) {
_truth[(i * Int(OpenCVTestCase.matSize) + j) * 3 + 0] = Float(i)
_truth[(i * Int(OpenCVTestCase.matSize) + j) * 3 + 1] = Float(j)
_truth[(i * Int(OpenCVTestCase.matSize) + j) * 3 + 2] = Float(i - j)
}
}
_truth[2] = 10000
try truth!.put(row: 0, col: 0, data: _truth)
try assertMatEqual(truth!, _3dPoints, OpenCVTestCase.EPS)
}
func testReprojectImageTo3DMatMatMatBooleanInt() throws {
let transformMatrix = Mat(rows: 4, cols: 4, type: CvType.CV_64F)
try transformMatrix.put(row: 0, col: 0, data: [0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1] as [Double])
let disparity = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_32F)
var disp = [Float](repeating: 0.0, count: Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize))
for i in 0..<Int(OpenCVTestCase.matSize) {
for j in 0..<Int(OpenCVTestCase.matSize) {
disp[i * Int(OpenCVTestCase.matSize) + j] = Float(i - j)
}
}
try disparity.put(row:0, col:0, data:disp)
let _3dPoints = Mat()
Calib3d.reprojectImageTo3D(disparity: disparity, _3dImage: _3dPoints, Q: transformMatrix, handleMissingValues: false, ddepth: CvType.CV_16S)
XCTAssertEqual(CvType.CV_16SC3, _3dPoints.type())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.rows())
XCTAssertEqual(OpenCVTestCase.matSize, _3dPoints.cols())
truth = Mat(rows: OpenCVTestCase.matSize, cols: OpenCVTestCase.matSize, type: CvType.CV_16SC3)
var _truth = [Int16](repeating: 0, count: Int(OpenCVTestCase.matSize * OpenCVTestCase.matSize * 3))
for i in 0..<Int(OpenCVTestCase.matSize) {
for j in 0..<Int(OpenCVTestCase.matSize) {
let start = (i * Int(OpenCVTestCase.matSize) + j) * 3
_truth[start + 0] = Int16(i)
_truth[start + 1] = Int16(j)
_truth[start + 2] = Int16(i - j)
}
}
try truth!.put(row: 0, col: 0, data: _truth)
try assertMatEqual(truth!, _3dPoints, OpenCVTestCase.EPS)
}
func testRodriguesMatMat() throws {
let r = Mat(rows: 3, cols: 1, type: CvType.CV_32F)
let R = Mat(rows: 3, cols: 3, type: CvType.CV_32F)
try r.put(row:0, col:0, data:[.pi, 0, 0] as [Float])
Calib3d.Rodrigues(src: r, dst: R)
truth = Mat(rows: 3, cols: 3, type: CvType.CV_32F)
try truth!.put(row:0, col:0, data:[1, 0, 0, 0, -1, 0, 0, 0, -1] as [Float])
try assertMatEqual(truth!, R, OpenCVTestCase.EPS)
let r2 = Mat()
Calib3d.Rodrigues(src: R, dst: r2)
try assertMatEqual(r, r2, OpenCVTestCase.EPS)
}
func testSolvePnPListOfPoint3ListOfPointMatMatMatMat() throws {
let intrinsics = Mat.eye(rows: 3, cols: 3, type: CvType.CV_64F)
try intrinsics.put(row: 0, col: 0, data: [400] as [Double])
try intrinsics.put(row: 1, col: 1, data: [400] as [Double])
try intrinsics.put(row: 0, col: 2, data: [640 / 2] as [Double])
try intrinsics.put(row: 1, col: 2, data: [480 / 2] as [Double])
let minPnpPointsNum: Int32 = 4
let points3d = MatOfPoint3f()
points3d.alloc(minPnpPointsNum)
let points2d = MatOfPoint2f()
points2d.alloc(minPnpPointsNum)
for i in 0..<minPnpPointsNum {
let x = Float.random(in: -50...50)
let y = Float.random(in: -50...50)
try points2d.put(row: i, col: 0, data: [x, y]) //add(Point(x, y))
try points3d.put(row: i, col: 0, data: [0, y, x]) // add(Point3(0, y, x))
}
let rvec = Mat()
let tvec = Mat()
Calib3d.solvePnP(objectPoints: points3d, imagePoints: points2d, cameraMatrix: intrinsics, distCoeffs: MatOfDouble(), rvec: rvec, tvec: tvec)
let truth_rvec = Mat(rows: 3, cols: 1, type: CvType.CV_64F)
try truth_rvec.put(row: 0, col: 0, data: [0, .pi / 2, 0] as [Double])
let truth_tvec = Mat(rows: 3, cols: 1, type: CvType.CV_64F)
try truth_tvec.put(row: 0, col: 0, data: [-320, -240, 400] as [Double])
try assertMatEqual(truth_rvec, rvec, OpenCVTestCase.EPS)
try assertMatEqual(truth_tvec, tvec, OpenCVTestCase.EPS)
}
func testComputeCorrespondEpilines() throws {
let fundamental = Mat(rows: 3, cols: 3, type: CvType.CV_64F)
try fundamental.put(row: 0, col: 0, data: [0, -0.577, 0.288, 0.577, 0, 0.288, -0.288, -0.288, 0])
let left = MatOfPoint2f()
left.alloc(1)
try left.put(row: 0, col: 0, data: [2, 3] as [Float]) //add(Point(x, y))
let lines = Mat()
let truth = Mat(rows: 1, cols: 1, type: CvType.CV_32FC3)
try truth.put(row: 0, col: 0, data: [-0.70735186, 0.70686162, -0.70588124])
Calib3d.computeCorrespondEpilines(points: left, whichImage: 1, F: fundamental, lines: lines)
try assertMatEqual(truth, lines, OpenCVTestCase.EPS)
}
func testConstants()
{
// calib3d.hpp: some constants have conflict with constants from 'fisheye' namespace
XCTAssertEqual(1, Calib3d.CALIB_USE_INTRINSIC_GUESS)
XCTAssertEqual(2, Calib3d.CALIB_FIX_ASPECT_RATIO)
XCTAssertEqual(4, Calib3d.CALIB_FIX_PRINCIPAL_POINT)
XCTAssertEqual(8, Calib3d.CALIB_ZERO_TANGENT_DIST)
XCTAssertEqual(16, Calib3d.CALIB_FIX_FOCAL_LENGTH)
XCTAssertEqual(32, Calib3d.CALIB_FIX_K1)
XCTAssertEqual(64, Calib3d.CALIB_FIX_K2)
XCTAssertEqual(128, Calib3d.CALIB_FIX_K3)
XCTAssertEqual(0x0800, Calib3d.CALIB_FIX_K4)
XCTAssertEqual(0x1000, Calib3d.CALIB_FIX_K5)
XCTAssertEqual(0x2000, Calib3d.CALIB_FIX_K6)
XCTAssertEqual(0x4000, Calib3d.CALIB_RATIONAL_MODEL)
XCTAssertEqual(0x8000, Calib3d.CALIB_THIN_PRISM_MODEL)
XCTAssertEqual(0x10000, Calib3d.CALIB_FIX_S1_S2_S3_S4)
XCTAssertEqual(0x40000, Calib3d.CALIB_TILTED_MODEL)
XCTAssertEqual(0x80000, Calib3d.CALIB_FIX_TAUX_TAUY)
XCTAssertEqual(0x100000, Calib3d.CALIB_USE_QR)
XCTAssertEqual(0x200000, Calib3d.CALIB_FIX_TANGENT_DIST)
XCTAssertEqual(0x100, Calib3d.CALIB_FIX_INTRINSIC)
XCTAssertEqual(0x200, Calib3d.CALIB_SAME_FOCAL_LENGTH)
XCTAssertEqual(0x400, Calib3d.CALIB_ZERO_DISPARITY)
XCTAssertEqual((1 << 17), Calib3d.CALIB_USE_LU)
XCTAssertEqual((1 << 22), Calib3d.CALIB_USE_EXTRINSIC_GUESS)
}
func testSolvePnPGeneric_regression_16040() throws {
let intrinsics = Mat.eye(rows: 3, cols: 3, type: CvType.CV_64F)
try intrinsics.put(row: 0, col: 0, data: [400] as [Double])
try intrinsics.put(row: 1, col: 1, data: [400] as [Double])
try intrinsics.put(row: 0, col: 2, data: [640 / 2] as [Double])
try intrinsics.put(row: 1, col: 2, data: [480 / 2] as [Double])
let minPnpPointsNum: Int32 = 4
let points3d = MatOfPoint3f()
points3d.alloc(minPnpPointsNum)
let points2d = MatOfPoint2f()
points2d.alloc(minPnpPointsNum)
for i in 0..<minPnpPointsNum {
let x = Float.random(in: -50...50)
let y = Float.random(in: -50...50)
try points2d.put(row: i, col: 0, data: [x, y]) //add(Point(x, y))
try points3d.put(row: i, col: 0, data: [0, y, x]) // add(Point3(0, y, x))
}
var rvecs = [Mat]()
var tvecs = [Mat]()
let rvec = Mat()
let tvec = Mat()
let reprojectionError = Mat(rows: 2, cols: 1, type: CvType.CV_64FC1)
Calib3d.solvePnPGeneric(objectPoints: points3d, imagePoints: points2d, cameraMatrix: intrinsics, distCoeffs: MatOfDouble(), rvecs: &rvecs, tvecs: &tvecs, useExtrinsicGuess: false, flags: .SOLVEPNP_IPPE, rvec: rvec, tvec: tvec, reprojectionError: reprojectionError)
let truth_rvec = Mat(rows: 3, cols: 1, type: CvType.CV_64F)
try truth_rvec.put(row: 0, col: 0, data: [0, .pi / 2, 0] as [Double])
let truth_tvec = Mat(rows: 3, cols: 1, type: CvType.CV_64F)
try truth_tvec.put(row: 0, col: 0, data: [-320, -240, 400] as [Double])
try assertMatEqual(truth_rvec, rvecs[0], 10 * OpenCVTestCase.EPS)
try assertMatEqual(truth_tvec, tvecs[0], 1000 * OpenCVTestCase.EPS)
}
func testGetDefaultNewCameraMatrixMat() {
let mtx = Calib3d.getDefaultNewCameraMatrix(cameraMatrix: gray0)
XCTAssertFalse(mtx.empty())
XCTAssertEqual(0, Core.countNonZero(src: mtx))
}
func testGetDefaultNewCameraMatrixMatSizeBoolean() {
let mtx = Calib3d.getDefaultNewCameraMatrix(cameraMatrix: gray0, imgsize: size, centerPrincipalPoint: true)
XCTAssertFalse(mtx.empty())
XCTAssertFalse(0 == Core.countNonZero(src: mtx))
// TODO_: write better test
}
func testUndistortMatMatMatMat() throws {
let src = Mat(rows: 3, cols: 3, type: CvType.CV_32F, scalar: Scalar(3))
let cameraMatrix = Mat(rows: 3, cols: 3, type: CvType.CV_32F)
try cameraMatrix.put(row: 0, col: 0, data: [1, 0, 1] as [Float])
try cameraMatrix.put(row: 1, col: 0, data: [0, 1, 2] as [Float])
try cameraMatrix.put(row: 2, col: 0, data: [0, 0, 1] as [Float])
let distCoeffs = Mat(rows: 1, cols: 4, type: CvType.CV_32F)
try distCoeffs.put(row: 0, col: 0, data: [1, 3, 2, 4] as [Float])
Calib3d.undistort(src: src, dst: dst, cameraMatrix: cameraMatrix, distCoeffs: distCoeffs)
truth = Mat(rows: 3, cols: 3, type: CvType.CV_32F)
try truth!.put(row: 0, col: 0, data: [0, 0, 0] as [Float])
try truth!.put(row: 1, col: 0, data: [0, 0, 0] as [Float])
try truth!.put(row: 2, col: 0, data: [0, 3, 0] as [Float])
try assertMatEqual(truth!, dst, OpenCVTestCase.EPS)
}
func testUndistortMatMatMatMatMat() throws {
let src = Mat(rows: 3, cols: 3, type: CvType.CV_32F, scalar: Scalar(3))
let cameraMatrix = Mat(rows: 3, cols: 3, type: CvType.CV_32F)
try cameraMatrix.put(row: 0, col: 0, data: [1, 0, 1] as [Float])
try cameraMatrix.put(row: 1, col: 0, data: [0, 1, 2] as [Float])
try cameraMatrix.put(row: 2, col: 0, data: [0, 0, 1] as [Float])
let distCoeffs = Mat(rows: 1, cols: 4, type: CvType.CV_32F)
try distCoeffs.put(row: 0, col: 0, data: [2, 1, 4, 5] as [Float])
let newCameraMatrix = Mat(rows: 3, cols: 3, type: CvType.CV_32F, scalar: Scalar(1))
Calib3d.undistort(src: src, dst: dst, cameraMatrix: cameraMatrix, distCoeffs: distCoeffs, newCameraMatrix: newCameraMatrix)
truth = Mat(rows: 3, cols: 3, type: CvType.CV_32F, scalar: Scalar(3))
try assertMatEqual(truth!, dst, OpenCVTestCase.EPS)
}
//undistortPoints(List<Point> src, List<Point> dst, Mat cameraMatrix, Mat distCoeffs)
func testUndistortPointsListOfPointListOfPointMatMat() {
let src = MatOfPoint2f(array: [Point2f(x: 1, y: 2), Point2f(x: 3, y: 4), Point2f(x: -1, y: -1)])
let dst = MatOfPoint2f()
let cameraMatrix = Mat.eye(rows: 3, cols: 3, type: CvType.CV_64FC1)
let distCoeffs = Mat(rows: 8, cols: 1, type: CvType.CV_64FC1, scalar: Scalar(0))
Calib3d.undistortPoints(src: src, dst: dst, cameraMatrix: cameraMatrix, distCoeffs: distCoeffs)
XCTAssertEqual(src.toArray(), dst.toArray())
}
}