public void setValues(ObjectPosRect objectPosRect)
{
classid = objectPosRect.classid;
confidence = objectPosRect.confidence;
x1 = objectPosRect.x1;
y1 = objectPosRect.y1;
x2 = objectPosRect.x2;
y2 = objectPosRect.y2;
}
//显示检测后的图像
public void ShowPicture(Mat frameMat, List <ObjectPosRect> objectPosRects, ObjectPosRect objectPosRect, long toltalMillis)
{
if (showPicture && mainForm != null && outPictureBox != null && mainForm.Visible)
{
if (frameMat != null)
{
foreach (ObjectPosRect rect in objectPosRects)
{
Cv2.Rectangle(frameMat, new OpenCvSharp.Point(rect.x1, rect.y1), new OpenCvSharp.Point(rect.x2, rect.y2), new Scalar(0, 255, 0), 2, LineTypes.Link4);
//string msg = (string)this.labelNames.GetValue(rect.classid - 1) + ", " + rect.confidence.ToString("F2") + ", " + toltalMillis;
string msg = (string)this.labelNames.GetValue(rect.classid) + ", " + rect.confidence.ToString("F2") + ", " + toltalMillis;
Cv2.PutText(frameMat, msg, new OpenCvSharp.Point(rect.x1, rect.y1 + 20), HersheyFonts.HersheySimplex, 0.95, new Scalar(0, 0, 255));
}
//计算对象的中心点，靠上方一些
int objCenterX = objectPosRect.x1 + (objectPosRect.x2 - objectPosRect.x1) / 2;
int objCenterY = objectPosRect.y1 + (objectPosRect.y2 - objectPosRect.y1) / 3;
objCenterX = detectionRect.x + objCenterX;
objCenterY = detectionRect.y + objCenterY;
//计算屏幕中心点
int screenCenterX = rawDetectionRect.x + rawDetectionRect.w / 2;
int screenCenterY = rawDetectionRect.y + rawDetectionRect.h / 2;
Cv2.Rectangle(frameMat, new OpenCvSharp.Point(objCenterX - 10 - detectionRect.x, objCenterY - 10 - detectionRect.y),
new OpenCvSharp.Point(objCenterX + 10 - detectionRect.x, objCenterY + 10 - detectionRect.y), new Scalar(0, 255, 0), 2, LineTypes.Link4);
Cv2.Rectangle(frameMat, new OpenCvSharp.Point(screenCenterX - 10 - detectionRect.x, screenCenterY - 10 - detectionRect.y),
new OpenCvSharp.Point(screenCenterX + 10 - detectionRect.x, screenCenterY + 10 - detectionRect.y), new Scalar(0, 0, 255), 2, LineTypes.Link4);
var frameBitmap = BitmapConverter.ToBitmap(frameMat);
outPictureBox.Image?.Dispose();
outPictureBox.Image = frameBitmap;
}
}
}
public long toltalMillis; //检测执行的时间
public void setValues(DetectionRect rawDetectionRect1, DetectionRect detectionRect1, Mat frameMat1, List <ObjectPosRect> objectPosRects1, ObjectPosRect maxConfidencePos1, int maxPersonW1, long toltalMillis1)
{
detectionResult4Rect = new DetectionResult4Rect();
detectionResult4Rect.rawDetectionRect = rawDetectionRect1;
detectionResult4Rect.detectionRect = detectionRect1;
detectionResult4Rect.maxConfidencePos = maxConfidencePos1;
detectionResult4Rect.maxPersonW = maxPersonW1;
frameMat = frameMat1;
objectPosRects = objectPosRects1;
toltalMillis = toltalMillis1;
}
//使用AI模型检测屏幕，识别人员及位置
public DetectionResult DetectionScreen()
{
long toltalMillis = 0;
long startTicks = DateTime.Now.Ticks;
//保存所有检测对象及最大置信度对象的位置
DetectionResult detectionResult = new DetectionResult();
List <ObjectPosRect> objectPosRects = new List <ObjectPosRect>();
ObjectPosRect maxConfidencePos = new ObjectPosRect();
maxConfidencePos.setValues(0, 0, 0, 0, 0, 0);
//拷贝需要检测屏幕区域的图像
//c#中处理图形，需要注意手动释放资源，及线程不安全的问题。
//System.Drawing.Bitmap bitmap = new Bitmap(this.detectionRect.w, this.detectionRect.h);
//System.Drawing.Graphics graphics = Graphics.FromImage(bitmap);
this.detectGraphics.CopyFromScreen(this.detectionRect.x, this.detectionRect.y, 0, 0, new System.Drawing.Size(this.detectionRect.w, this.detectionRect.h), CopyPixelOperation.SourceCopy);
try
{
//注意这里不用using语句，因为会隐式释放图像对象，导致外面其他函数使用对象时无法访问。
Mat frameMat = BitmapConverter.ToMat(this.detectBitmap);
if (frameMat != null && !frameMat.Empty())
{
//屏幕截图和视频截图格式不一样，需要进行图像格式转换
Cv2.CvtColor(frameMat, frameMat, ColorConversionCodes.RGBA2RGB);
// Convert Mat to batch of images
//var frameWidth = frameMat.Cols;
//var frameHeight = frameMat.Rows;
var frameWidth = 512;
var frameHeight = 512;
//注意转换位输入数据时，最好保持原图大小，或者按比例缩放，相对于不按比例缩放，可以让识别准确率大幅提升。
//注意scaleFactor和scalemean参数设置，涉及模型输入数据归一化，严重影响模型准确率。
//using (var inputBlob = CvDnn.BlobFromImage(frameMat, 0.008, new OpenCvSharp.Size(320, 320), new Scalar(104, 117, 123), true, false))
//using (var inputBlob = CvDnn.BlobFromImage(frameMat, 0.008, new OpenCvSharp.Size(320, 320), new Scalar(103.939, 116.779, 123.68), true, false))
//using (var inputBlob = CvDnn.BlobFromImage(frameMat, 0.008, new OpenCvSharp.Size(frameWidth, frameHeight), new Scalar(103.939, 116.779, 123.68), true, false))
//using (var inputBlob = CvDnn.BlobFromImage(frameMat, 1.0 / 127.5, new OpenCvSharp.Size(frameWidth, frameHeight), new Scalar(127.5, 127.5, 127.5), true, false))
using (var inputBlob = CvDnn.BlobFromImage(frameMat, 1.0 / 255, new OpenCvSharp.Size(frameWidth, frameHeight), new Scalar(123.675, 116.28, 103.53), true, false))
{
//使用AI模型推理检测图像
detectionNet.SetInput(inputBlob);
var output = detectionNet.Forward();
//分析检测结果
var detectionMat = new Mat(output.Size(2), output.Size(3), MatType.CV_32F, output.Ptr(0));
for (int i = 0; i < detectionMat.Rows; i++)
{
float confidence = detectionMat.At <float>(i, 2);
if (confidence > 0.30)
{
int classid = (int)detectionMat.At <float>(i, 1);
//判断识别的类型，这个应用场景只显示识别的人，提升处理效率
//if (classid < this.labelNames.Length && (classid == 1 || classid != 1))
//if (classid < this.labelNames.Length && (classid == 1))
if (classid < this.labelNames.Length && (classid == 0))
{
int x1 = (int)(detectionMat.At <float>(i, 3) * frameWidth);
int y1 = (int)(detectionMat.At <float>(i, 4) * frameHeight);
int x2 = (int)(detectionMat.At <float>(i, 5) * frameWidth);
int y2 = (int)(detectionMat.At <float>(i, 6) * frameHeight);
ObjectPosRect objectPosRect = new ObjectPosRect();
objectPosRect.setValues(classid, confidence, x1, y1, x2, y2);
objectPosRects.Add(objectPosRect);
//判断是否是最大人员模型；
//为避免处理逻辑太复杂，最大模型宽度设置为一个常量；
//if ((objectPosRect.x2 - objectPosRect.x1) < this.detectionRect.w / 3 &&
// (this.maxPersonPos.x2 - this.maxPersonPos.x1 < objectPosRect.x2 - objectPosRect.x1))
// this.maxPersonPos.setValues(objectPosRect);
//为保障项目，排除太大或者太小的模型
if ((objectPosRect.x2 - objectPosRect.x1) <= 200 && (objectPosRect.x2 - objectPosRect.x1) >= 10 &&
(objectPosRect.y2 - objectPosRect.y1) <= 280 && (objectPosRect.y2 - objectPosRect.y1) >= 10)
{
//判断是否是游戏操作人,模型位置为屏幕游戏者位置
//游戏者的位置在屏幕下方靠左一点，大概 860/1920处
//另外游戏中左右摇摆幅度较大，所以x轴的兼容值要设置大一些。
if (Math.Abs(objectPosRect.x1 + (objectPosRect.x2 - objectPosRect.x1) / 2 - this.playerCentX) <= 100 &&
objectPosRect.y1 > this.detectionRect.h * 1 / 2 &&
Math.Abs(this.detectionRect.h - objectPosRect.y2) <= 10)
{
//排除游戏者自己
//var testi = 0;
}
else
{
//保存置信度最大的人员的位置,同时排除屏幕下方正中的游戏者自己
if (objectPosRect.confidence > maxConfidencePos.confidence)
{
maxConfidencePos.setValues(objectPosRect);
}
}
}
long endTicks = DateTime.Now.Ticks;
toltalMillis = (endTicks - startTicks) / 10000;
}
}
}
}
detectionResult.setValues(this.rawDetectionRect, this.detectionRect, frameMat, objectPosRects, maxConfidencePos, this.maxPersonW, toltalMillis);
//图像显示放到其他线程中处理
//this.ShowPicture(frameMat, objectPosRects, maxConfidencePos, toltalMillis);
}
//Thread.Sleep(100);
}
catch (Exception ex)
{
Console.WriteLine(ex.Message);
}
//使用类属性，只创建了一个对象，不能手工释放
//graphics.Dispose();
//bitmap.Dispose();
return(detectionResult);
}