104 lines
4.1 KiB
C#
104 lines
4.1 KiB
C#
using OpenCVForUnity.CoreModule;
|
|
using OpenCVForUnity.ImgprocModule;
|
|
using OpenCVForUnity.MlModule;
|
|
using OpenCVForUnity.UnityUtils;
|
|
using UnityEngine;
|
|
using UnityEngine.SceneManagement;
|
|
|
|
namespace OpenCVForUnityExample
|
|
{
|
|
/// <summary>
|
|
/// KNN Example
|
|
/// An example to understand the concepts of the k-Nearest Neighbour (kNN) algorithm.
|
|
/// https://docs.opencv.org/4.x/d5/d26/tutorial_py_knn_understanding.html
|
|
/// </summary>
|
|
public class KNNExample : MonoBehaviour
|
|
{
|
|
// Use this for initialization
|
|
void Start()
|
|
{
|
|
//if true, The error log of the Native side OpenCV will be displayed on the Unity Editor Console.
|
|
Utils.setDebugMode(true);
|
|
|
|
|
|
// Feature set containing (x,y) values of 25 known/training data
|
|
Mat trainData = new Mat(25, 2, CvType.CV_32FC1);
|
|
using (Mat trainDataInt = new Mat(25, 2, CvType.CV_16SC1))
|
|
{
|
|
Core.randu(trainDataInt, 0, 100); // random values
|
|
trainDataInt.convertTo(trainData, CvType.CV_32FC1);
|
|
}
|
|
//Debug.Log(trainData.dump());
|
|
|
|
// Label each one either Red or Blue with numbers 0 and 1
|
|
Mat responses = new Mat(25, 1, CvType.CV_32FC1);
|
|
using (Mat responsesInt = new Mat(25, 1, CvType.CV_16SC1))
|
|
{
|
|
Core.randu(responsesInt, 0, 2); // random values
|
|
responsesInt.convertTo(responses, CvType.CV_32FC1);
|
|
}
|
|
//Debug.Log(responses.dump());
|
|
|
|
|
|
KNearest knn = KNearest.create();
|
|
knn.train(trainData, Ml.ROW_SAMPLE, responses);
|
|
|
|
Mat newcomer = new Mat(1, 2, CvType.CV_32FC1, new Scalar(50, 50));
|
|
Mat results = new Mat();
|
|
Mat neighbours = new Mat();
|
|
Mat dist = new Mat();
|
|
knn.findNearest(newcomer, 3, results, neighbours, dist);
|
|
|
|
|
|
Mat plotMat = new Mat(500, 500, CvType.CV_8UC4, new Scalar(255, 255, 255, 255));
|
|
|
|
// Take Red neighbours and plot them
|
|
// Take Blue neighbours and plot them
|
|
for (int i = 0; i < trainData.rows(); i++)
|
|
{
|
|
bool red = ((int)responses.get(i, 0)[0] == 0);
|
|
|
|
double x = trainData.get(i, 0)[0];
|
|
double y = trainData.get(i, 1)[0];
|
|
|
|
Imgproc.circle(plotMat, new Point(x * 5f, y * 5f), 5, red ? new Scalar(255, 0, 0, 255) : new Scalar(0, 0, 255, 255), -1);
|
|
}
|
|
// Plot newcomer and the neighbours distance circle
|
|
Imgproc.circle(plotMat, new Point(50f * 5f, 50f * 5f), 5, new Scalar(0, 255, 0, 255), -1);
|
|
Imgproc.circle(plotMat, new Point(50f * 5f, 50f * 5f), (int)(Mathf.Sqrt((float)dist.get(0, 2)[0]) * 5f), new Scalar(0, 255, 0, 255), 1);
|
|
|
|
|
|
Debug.Log("0:Red / 1:Blue");
|
|
Debug.Log("result: " + results.dump());
|
|
Debug.Log("neighbours: " + neighbours.dump());
|
|
Debug.Log("distance: " + dist.dump());
|
|
|
|
Imgproc.putText(plotMat, "0:Red / 1:Blue", new Point(5, 30), Imgproc.FONT_HERSHEY_SIMPLEX, 1.0, new Scalar(0, 0, 0, 255));
|
|
Imgproc.putText(plotMat, "result: " + results.dump(), new Point(5, 65), Imgproc.FONT_HERSHEY_SIMPLEX, 1.0, new Scalar(0, 0, 0, 255));
|
|
Imgproc.putText(plotMat, "neighbours: " + neighbours.dump(), new Point(5, 100), Imgproc.FONT_HERSHEY_SIMPLEX, 1.0, new Scalar(0, 0, 0, 255));
|
|
Imgproc.putText(plotMat, "distance: " + dist.dump(), new Point(5, 135), Imgproc.FONT_HERSHEY_SIMPLEX, 1.0, new Scalar(0, 0, 0, 255));
|
|
|
|
|
|
Texture2D texture = new Texture2D(plotMat.cols(), plotMat.rows(), TextureFormat.RGBA32, false);
|
|
Utils.matToTexture2D(plotMat, texture);
|
|
gameObject.GetComponent<Renderer>().material.mainTexture = texture;
|
|
|
|
|
|
Utils.setDebugMode(false);
|
|
}
|
|
|
|
// Update is called once per frame
|
|
void Update()
|
|
{
|
|
|
|
}
|
|
|
|
/// <summary>
|
|
/// Raises the back button click event.
|
|
/// </summary>
|
|
public void OnBackButtonClick()
|
|
{
|
|
SceneManager.LoadScene("OpenCVForUnityExample");
|
|
}
|
|
}
|
|
} |