Skip to content

Latest commit

 

History

History
121 lines (91 loc) · 4.2 KB

README.md

File metadata and controls

121 lines (91 loc) · 4.2 KB

Video4j - Facedetect

This project contains APIs that provide access to different Face detection implementations via the Video4j library.

Status

There are still some kinks which need to be worked out. Thus the library has not yet been published. The API may change in the future.

Example

ezgif-2-66ea4664a0

Video4j.init();
DLibFacedetector detector = DLibFacedetector.create();
detector.setMinFaceHeightFactor(0.01f);
detector.enableCNNDetector();
detector.enableLandmarks();
detector.enableLandmarks();

try (Video video = Videos.open("src/test/resources/pexels-mikhail-nilov-7626566.mp4")) {
	FacedetectorMetrics metrics = FacedetectorMetrics.create();
	Stream<FaceVideoFrame> frameStream = video.streamFrames()
		.filter(frame -> {
			return frame.number() % 5 == 0;
		})
		.map(frame -> {
			CVUtils.boxFrame2(frame, 384);
			return frame;
		})
		.map(detector::detect)
		.filter(FaceVideoFrame::hasFace)
		.map(metrics::track)
		.map(detector::markFaces)
		.map(detector::markLandmarks)
		.map(frame -> detector.drawMetrics(frame, metrics, new Point(25, 45)))
		.map(frame -> cropToFace(frame, 0));
	VideoUtils.showVideoFrameStream(frameStream);
}

Available Detectors

OpenCV

The OpenCV classifier based face detection needs to be initialized before usage.

// Face detection classifiers
CVFacedetection.loadLbpcascade();
CVFacedetection.loadHaarcascade();

// Landmark detection models
CVFacedetection.loadLBFLandmarkModel();
CVFacedetection.loadKazemiLandmarkModel();

// Open video and load frames
See Video4j API on how to handle videos

// Run face detection 
// The landmark detection can be turned of if not needed. In this case no landmark model is needed.
FaceVideoFrame faceframe = CVFacedetection.scan(frame, 0.05f, true);

dlib

The face detector implementation loads the needed models automatically. At the moment two options are available:

  • HOG face detector
  • CNN face detector which can utilize GPU
// Open video and load frames
…

// Run the face detection using dlib
boolean useCNN = true;
boolean loadEmbeddings = true;
FaceVideoFrame faceframe = DLibFacedetection.scan(frame, 0.05f, useCNN, loadEmbeddings);

Face Extraction data

faceframe.hasFace(); // Check if the frame contains a detected face
List<Face> faces = faceframe.faces(); // Access the faces
Face face = faces.get(0);
face.start(); // Upper left point of the face
face.end(); // Lower right point of the face
Dimension dim = face.dimension(); // Dimension of the face area in pixel
List<Point> landmarks = face.getLandmarks(); // Load the detected landmarks
face.getEmbeddings(); // Access the embeddings vector data

Model sources

dlib:

OpenCV:

Test footage sources