I'm exploring Microsoft Cognitive Face API and I'm very new to it. I'm able to achieve Face Attributes with an image which is easy but, my question is how can I get Face Attributes of a person in a real time video feed from Kinect in WPF c#. It would be great if somebody can help me out. Thanks in Advance!
I have tried capturing the frame from the Kinect color feed at every 2 seconds to some file location and use that file path and converted it to a Stream and then pass it on to the Face-API Functions and that worked. Following is the code I tried.
namespace CognitiveFaceAPISample
{
public partial class MainWindow : Window
{
private readonly IFaceServiceClient faceServiceClient = new FaceServiceClient("c2446f84b1eb486ca11e2f5d6e670878");
KinectSensor ks;
ColorFrameReader cfr;
byte[] colorData;
ColorImageFormat format;
WriteableBitmap wbmp;
BitmapSource bmpSource;
int imageSerial;
DispatcherTimer timer,timer2;
string streamF = "Frames//frame.jpg";
public MainWindow()
{
InitializeComponent();
ks = KinectSensor.GetDefault();
ks.Open();
var fd = ks.ColorFrameSource.CreateFrameDescription(ColorImageFormat.Bgra);
uint frameSize = fd.BytesPerPixel * fd.LengthInPixels;
colorData = new byte[frameSize];
format = ColorImageFormat.Bgra;
imageSerial = 0;
cfr = ks.ColorFrameSource.OpenReader();
cfr.FrameArrived += cfr_FrameArrived;
}
void cfr_FrameArrived(object sender, ColorFrameArrivedEventArgs e)
{
if (e.FrameReference == null) return;
using (ColorFrame cf = e.FrameReference.AcquireFrame())
{
if (cf == null) return;
cf.CopyConvertedFrameDataToArray(colorData, format);
var fd = cf.FrameDescription;
// Creating BitmapSource
var bytesPerPixel = (PixelFormats.Bgr32.BitsPerPixel) / 8;
var stride = bytesPerPixel * cf.FrameDescription.Width;
bmpSource = BitmapSource.Create(fd.Width, fd.Height, 96.0, 96.0, PixelFormats.Bgr32, null, colorData, stride);
// WritableBitmap to show on UI
wbmp = new WriteableBitmap(bmpSource);
FacePhoto.Source = wbmp;
}
}
private void SaveImage(BitmapSource image)
{
try
{
FileStream stream = new System.IO.FileStream(@"Frames\frame.jpg", System.IO.FileMode.OpenOrCreate);
JpegBitmapEncoder encoder = new JpegBitmapEncoder();
encoder.FlipHorizontal = true;
encoder.FlipVertical = false;
encoder.QualityLevel = 30;
encoder.Frames.Add(BitmapFrame.Create(image));
encoder.Save(stream);
stream.Close();
}
catch (Exception)
{
}
}
private void Window_Loaded(object sender, RoutedEventArgs e)
{
timer = new DispatcherTimer { Interval = TimeSpan.FromSeconds(2) };
timer.Tick += Timer_Tick;
timer.Start();
timer2 = new DispatcherTimer { Interval = TimeSpan.FromSeconds(5) };
timer2.Tick += Timer2_Tick;
timer2.Start();
}
private void Timer_Tick(object sender, EventArgs e)
{
SaveImage(bmpSource);
}
private async void Timer2_Tick(object sender, EventArgs e)
{
Title = "Detecting...";
FaceRectangle[] faceRects = await UploadAndDetectFaces(streamF);
Face[] faceAttributes = await UploadAndDetectFaceAttributes(streamF);
Title = String.Format("Detection Finished. {0} face(s) detected", faceRects.Length);
if (faceRects.Length > 0)
{
DrawingVisual visual = new DrawingVisual();
DrawingContext drawingContext = visual.RenderOpen();
drawingContext.DrawImage(bmpSource,
new Rect(0, 0, bmpSource.Width, bmpSource.Height));
double dpi = bmpSource.DpiX;
double resizeFactor = 96 / dpi;
foreach (var faceRect in faceRects)
{
drawingContext.DrawRectangle(
Brushes.Transparent,
new Pen(Brushes.Red, 2),
new Rect(
faceRect.Left * resizeFactor,
faceRect.Top * resizeFactor,
faceRect.Width * resizeFactor,
faceRect.Height * resizeFactor
)
);
}
drawingContext.Close();
RenderTargetBitmap faceWithRectBitmap = new RenderTargetBitmap(
(int)(bmpSource.PixelWidth * resizeFactor),
(int)(bmpSource.PixelHeight * resizeFactor),
96,
96,
PixelFormats.Pbgra32);
faceWithRectBitmap.Render(visual);
FacePhoto.Source = faceWithRectBitmap;
}
if (faceAttributes.Length > 0)
{
foreach (var faceAttr in faceAttributes)
{
Label lb = new Label();
//Canvas.SetLeft(lb, lb.Width);
lb.Content = faceAttr.FaceAttributes.Gender;// + " " + faceAttr.Gender + " " + faceAttr.FacialHair + " " + faceAttr.Glasses + " " + faceAttr.HeadPose + " " + faceAttr.Smile;
lb.FontSize = 50;
lb.Width = 200;
lb.Height = 100;
stack.Children.Add(lb);
}
}
}
private async Task<FaceRectangle[]> UploadAndDetectFaces(string imageFilePath)
{
try
{
using (Stream imageFileStream = File.OpenRead(imageFilePath))
{
var faces = await faceServiceClient.DetectAsync(imageFilePath);
var faceRects = faces.Select(face => face.FaceRectangle);
var faceAttrib = faces.Select(face => face.FaceAttributes);
return faceRects.ToArray();
}
}
catch (Exception)
{
return new FaceRectangle[0];
}
}
private async Task<Face[]> UploadAndDetectFaceAttributes(string imageFilePath)
{
try
{
using (Stream imageFileStream = File.Open(imageFilePath, FileMode.Open, FileAccess.Read, FileShare.ReadWrite))
{
var faces = await faceServiceClient.DetectAsync(imageFileStream, true, true, new FaceAttributeType[] { FaceAttributeType.Gender, FaceAttributeType.Age, FaceAttributeType.Smile, FaceAttributeType.Glasses, FaceAttributeType.HeadPose, FaceAttributeType.FacialHair });
return faces.ToArray();
}
}
catch (Exception)
{
return new Face[0];
}
}
}
Above code worked well. But, I want to convert each frame of Kinect Color Feed directly to Stream and I have no idea how to do it though I searched but nothing worked for me. If Somebody can help me then it'll be great. Thanks!
Instead of persisting the frame to a file in SaveImage, you can persist it to a MemoryStream, rewind it (by calling Position = 0), and send that stream to DetectAsync().
Also note that in UploadAndDetectFaces, you should send imageFileStream, not imageFilePath, to DetectAsync(). You probably don't want to call both UploadAndDetectFaces and UploadAndDetectFaceAttributes anyway, since you're just doubling your work (and quota/rate-limit hit.)