在Android服务中运行OpenCV眼睛检测 [英] Running OpenCV eye detection from within Android service
本文介绍了在Android服务中运行OpenCV眼睛检测的处理方法，对大家解决问题具有一定的参考价值，需要的朋友们下面随着小编来一起学习吧！
问题描述
我想通过OpenCV4Android从Android后台服务运行眼睛检测.我有一段代码可以很好地运行，但是作为Activity而不是服务.我了解Android相机必须先打开预览.因此，我创建了一个预览(要使它隐藏起来，可以制作一个小的预览，因为我希望处理在后台进行)，并启动了摄像头进行录制.摄像头成功启动，但是OpenCV无法检测到眼睛和面部.它仅加载xml分类器.我希望打开相机进行录制时会调用诸如onCameraViewStarted和onCameraFrame之类的OpenCV回调，但是没有.
I want to run eye detection by OpenCV4Android from Android background service. I have a piece of code that runs well but as an Activity not service. I understand that the Android camera must have a preview to open. So I have created a preview (small one to make it looks hidden, since I want the processing to be in the background) and started the camera for recording. The camera starts successfully, but OpenCV doesn't detect eyes and faces. It only loads the xml classifiers. I expected the callbacks of OpenCV like onCameraViewStarted and onCameraFrame to get called when I open the camera for recording, but they didn't.
这是代码:
public class BackgroundService extends Service implements SurfaceHolder.Callback, CameraBridgeViewBase.CvCameraViewListener2 {
private static final String TAG = "OCVSample::Activity";
private static final Scalar FACE_RECT_COLOR = new Scalar(0, 255, 0, 255);
public static final int JAVA_DETECTOR = 0;
private static final int TM_SQDIFF = 0;
private static final int TM_SQDIFF_NORMED = 1;
private static final int TM_CCOEFF = 2;
private static final int TM_CCOEFF_NORMED = 3;
private static final int TM_CCORR = 4;
private static final int TM_CCORR_NORMED = 5;
private int learn_frames = 0;
private Mat templateR;//right eye template
private Mat templateL; // left eye template
int method = 0;
private MenuItem mItemFace50;
private MenuItem mItemFace40;
private MenuItem mItemFace30;
private MenuItem mItemFace20;
private MenuItem mItemType;
private Mat mRgba;
private Mat mGray;
// matrix for zooming
private Mat mZoomWindow;
private Mat mZoomWindow2;
private File mCascadeFile;
private CascadeClassifier mJavaDetector;
private CascadeClassifier mJavaDetectorEye;
private int mDetectorType = JAVA_DETECTOR;
private String[] mDetectorName;
private float mRelativeFaceSize = 0.2f;
private int mAbsoluteFaceSize = 0;
private CameraBridgeViewBase mOpenCvCameraView;
private SeekBar mMethodSeekbar;
private TextView mValue;
double xCenter = -1;
double yCenter = -1;
MediaRecorder mediaRecorder;
// Binder given to clients
private final IBinder mBinder = new LocalBinder();
public class LocalBinder extends Binder {
BackgroundService getService() {
// Return this instance of this service so clients can call public methods
return BackgroundService.this;
}
}//end inner class that returns an instance of the service.
@Override
public IBinder onBind(Intent intent) {
return mBinder;
}//end onBind.
private WindowManager windowManager;
private SurfaceView surfaceView;
private Camera camera = null;
@Override
public void onCreate() {
// Start foreground service to avoid unexpected kill
Notification notification = new Notification.Builder(this)
.setContentTitle("Background Video Recorder")
.setContentText("")
.setSmallIcon(R.drawable.vecsat_logo)
.build();
startForeground(1234, notification);
// Create new SurfaceView, set its size to 1x1, move it to the top left corner and set this service as a callback
windowManager = (WindowManager) this.getSystemService(Context.WINDOW_SERVICE);
surfaceView = new SurfaceView(this);
WindowManager.LayoutParams layoutParams = new WindowManager.LayoutParams(
100, 100,
WindowManager.LayoutParams.TYPE_SYSTEM_OVERLAY,
WindowManager.LayoutParams.FLAG_WATCH_OUTSIDE_TOUCH,
PixelFormat.TRANSLUCENT
);
Log.i(TAG, "100 x 100 executed");
layoutParams.gravity = Gravity.LEFT | Gravity.TOP;
windowManager.addView(surfaceView, layoutParams);
surfaceView.getHolder().addCallback(this);
//constructor:
mDetectorName = new String[2];// contains 3 positions..
mDetectorName[JAVA_DETECTOR] = "Java"; //let the detector be of type java detector, specify that in the JAVA_DETECTOR index.
Log.i(TAG, "Instantiated new " + ((Object) this).getClass().getSimpleName());
OpenCVLoader.initAsync(OpenCVLoader.OPENCV_VERSION_2_4_11, this,
mLoaderCallback); //once the application is resumed reload the library.
}
// Method called right after Surface created (initializing and starting MediaRecorder)
@Override
public void surfaceCreated(SurfaceHolder surfaceHolder) {
Log.i(TAG, "surfaceCreated method");
camera = Camera.open(1);
camera.unlock();
mediaRecorder = new MediaRecorder();
mediaRecorder.setPreviewDisplay(surfaceHolder.getSurface());
mediaRecorder.setCamera(camera);
mediaRecorder.setAudioSource(MediaRecorder.AudioSource.CAMCORDER);
mediaRecorder.setVideoSource(MediaRecorder.VideoSource.CAMERA);
mediaRecorder.setProfile(CamcorderProfile.get(CamcorderProfile.QUALITY_HIGH));
mediaRecorder.setOutputFile(
Environment.getExternalStorageDirectory()+"/"+
DateFormat.format("yyyy-MM-dd_kk-mm-ss", new Date().getTime())+
".mp4"
);
try { mediaRecorder.prepare(); } catch (Exception e) {}
mediaRecorder.start();
}
// Stop recording and remove SurfaceView
@Override
public void onDestroy() {
Log.i(TAG, "surfaceDestroyed method");
camera.lock();
camera.release();
windowManager.removeView(surfaceView);
}
@Override
public void surfaceChanged(SurfaceHolder surfaceHolder, int format, int width, int height) {}
@Override
public void surfaceDestroyed(SurfaceHolder surfaceHolder) {
}
private BaseLoaderCallback mLoaderCallback = new BaseLoaderCallback(this) {
@Override
public void onManagerConnected(int status) {
//int status, status of initialization, sucess or not..
//now make a switch for the status cases: under success case do the work, load the classifiers..
switch (status) {
case LoaderCallbackInterface.SUCCESS: {
Log.i(TAG, "OpenCV loaded successfully"); // was loaded and initialized successfully..
try {
// load cascade file from application resources
InputStream is = getResources().openRawResource(
R.raw.lbpcascade_frontalface); // get the face classifier from the resource.
File cascadeDir = getDir("cascade", Context.MODE_PRIVATE);
mCascadeFile = new File(cascadeDir,
"lbpcascade_frontalface.xml"); // create a directory inside your app, and a file inside it to store the
FileOutputStream os = new FileOutputStream(mCascadeFile); // prepare an output stream that will write the classifier's code on the file in the app.
//read and write
byte[] buffer = new byte[4096];
int bytesRead;
while ((bytesRead = is.read(buffer)) != -1) {
os.write(buffer, 0, bytesRead);
}
is.close();
os.close();
// --------------------------------- load left eye
// classificator -----------------------------------
InputStream iser = getResources().openRawResource(
R.raw.haarcascade_lefteye_2splits);
File cascadeDirER = getDir("cascadeER",
Context.MODE_PRIVATE);
File cascadeFileER = new File(cascadeDirER,
"haarcascade_eye_right.xml");
FileOutputStream oser = new FileOutputStream(cascadeFileER);
byte[] bufferER = new byte[4096];
int bytesReadER;
while ((bytesReadER = iser.read(bufferER)) != -1) {
oser.write(bufferER, 0, bytesReadER);
}
iser.close();
oser.close();
//check if you can load the classifer.
mJavaDetector = new CascadeClassifier(
mCascadeFile.getAbsolutePath());
if (mJavaDetector.empty()) {
Toast.makeText(getApplicationContext(), "face classifier error", Toast.LENGTH_LONG).show();
Log.e(TAG, "Failed to load cascade face classifier");
mJavaDetector = null;
} else
Log.i(TAG, "Loaded cascade classifier from "
+ mCascadeFile.getAbsolutePath());
mJavaDetectorEye = new CascadeClassifier(
cascadeFileER.getAbsolutePath());
if (mJavaDetectorEye.empty()) {
Toast.makeText(getApplicationContext(), "eye classifer error", Toast.LENGTH_LONG).show();
Log.e(TAG, "Failed to load cascade eye classifier");
mJavaDetectorEye = null;
} else
Log.i(TAG, "Loaded cascade classifier from "
+ mCascadeFile.getAbsolutePath());
cascadeDir.delete();
} catch (IOException e) {
e.printStackTrace();
Log.e(TAG, "Failed to load cascade. Exception thrown: " + e);
}
//Whether classifiers are opened or not, open the front camera.
// mOpenCvCameraView.setCameraIndex(1);
//mOpenCvCameraView.enableFpsMeter(); // What is this? This method enables label with fps value on the screen
// mOpenCvCameraView.enableView(); // What? This means enable connecting to the camera.
}
break;
default: {
//When the loading of the libarary is failed
super.onManagerConnected(status);
}
break;
}
}
}; // end the class.
public void onCameraViewStarted(int width, int height) {
Log.i(TAG, "onCameraViewStarted method");
//onCameraViewStarted callback will be delivered only after enableView is called and surface is available
//This method is a member of CvCameraViewListener2, and we must implement it.
mGray = new Mat(); //initialize new gray scale matrix to contain the img pixels.
mRgba = new Mat(); //initialize new rgb matrix to contain the img pixels.
}
public void onCameraViewStopped() {
Log.i(TAG, "onCameraViewStopped method");
//Release the allocated memory
//release the matrix, this releases the allocated space in memory, since mat contains a header that contains img info and a pointer that points to the matrix in the memory.
mGray.release();
mRgba.release();
mZoomWindow.release();
mZoomWindow2.release();
}
public Mat onCameraFrame(CameraBridgeViewBase.CvCameraViewFrame inputFrame) {
Log.i(TAG, "onCameraFrame method");
//This method is a member of CvCameraViewListener2, and we must implement it.
// In this method we get every frame from the camera and process it in order to track the objects.
//inputFrame is the received frame from the camera.
mRgba = inputFrame.rgba(); //convert the frame to rgba scale, then assign this value to the rgba Mat img matrix.
mGray = inputFrame.gray(); //convert the frame to gray scale, then assign this value to the gray Mat img matrix.
//Shall we consider Flipping the camera img horizontally.
if (mAbsoluteFaceSize == 0) {
int height = mGray.rows(); //get the height of the captured frame stored in mgray Mat array (rows), why gray to rgb???
if (Math.round(height * mRelativeFaceSize) > 0) { //multiply that height with 0.2... Is the result > 0?
//if yes this indicates that there is a frame that was captured (it's height is not zero), so set the face size to
// Math.round(height * mRelativeFaceSize)
mAbsoluteFaceSize = Math.round(height * mRelativeFaceSize);
}
}
if (mZoomWindow == null || mZoomWindow2 == null)
CreateAuxiliaryMats();
MatOfRect faces = new MatOfRect(); //a matrix that will contain rectangles around the face (including the faces inside the rectangles), it will be filled by detectMultiScale method.
//if mJavaDetector is not null, this contains the face classifier that we have loaded previously
if (mJavaDetector != null)
//if not null, use this classifier to detect faces.
mJavaDetector.detectMultiScale(mGray, faces, 1.1, 2,
2, // TODO: objdetect.CV_HAAR_SCALE_IMAGE
new Size(mAbsoluteFaceSize, mAbsoluteFaceSize),
new Size());
//in th function detectMultiScale above,
// faces is the array that will contain the rectangles around the detected face.
// the 3rd param: specifies how much the image size is reduced at each image scale.
//4th param: Parameter specifying how many neighbors each candidate rectangle should have to retain it.
//5: :)
//6: Minimum possible object size. Objects smaller than that are ignored (if you set a very small minimum value, your app will run heavily).
//7: Maximum possible object size. Objects larger than that are ignored. Both minimum and maximum should be set carefully to avoid slow running of the app.
Rect[] facesArray = faces.toArray(); //array of faces
for (int i = 0; i < facesArray.length; i++) {
/* Imgproc.rectangle(mRgba, facesArray[i].tl(), facesArray[i].br(),
FACE_RECT_COLOR, 3);*/
//Now draw rectangles around the obtained faces, and a circle at each rectangle center.
//mrgba in the line bellow means that the rectangle should be drawn on the colored img.
//facesArray[i].tl() returns a Point: Template class for 2D points specified by its coordinates x and y -> Template class
// facesArray[i].x and facesArray[i].y are the x and y coords of the top left top corner.
Core.rectangle(mRgba, facesArray[i].tl(), facesArray[i].br(), FACE_RECT_COLOR, 3);
//calculate the center in x and y coords.
xCenter = (facesArray[i].x + facesArray[i].width + facesArray[i].x) / 2;
yCenter = (facesArray[i].y + facesArray[i].y + facesArray[i].height) / 2;
Point center = new Point(xCenter, yCenter); //store the center.
//Imgproc.circle(mRgba, center, 10, new Scalar(255, 0, 0, 255), 3);
Core.circle(mRgba, center, 10, new Scalar(255, 0, 0, 255), 3); //draw a red circle at the center of the face rectangle.
/*Imgproc.putText(mRgba, "[" + center.x + "," + center.y + "]",
new Point(center.x + 20, center.y + 20),
Core.FONT_HERSHEY_SIMPLEX, 0.7, new Scalar(255, 255, 255,
255));*/
//write the coordinates of the rectangle center:
Core.putText(mRgba, "[" + center.x + "," + center.y + "]",
new Point(center.x + 20, center.y + 20) , // this is the bottom left corner of the text string
Core.FONT_HERSHEY_SIMPLEX, 0.7, new Scalar(255, 255, 255,
255));
Rect r = facesArray[i]; //get the currect face, we want to use it to detect the eyes inside it.
// compute the eye area
//Rect (x, y, w, h)
Rect eyearea = new Rect(r.x + r.width / 8,
(int) (r.y + (r.height / 4.5)), r.width - 2 * r.width / 8,
(int) (r.height / 3.0));
// split it
Rect eyearea_right = new Rect(r.x + r.width / 16,
(int) (r.y + (r.height / 4.5)),
(r.width - 2 * r.width / 16) / 2, (int) (r.height / 3.0));
Rect eyearea_left = new Rect(r.x + r.width / 16
+ (r.width - 2 * r.width / 16) / 2,
(int) (r.y + (r.height / 4.5)),
(r.width - 2 * r.width / 16) / 2, (int) (r.height / 3.0));
// draw the area - mGray is working grayscale mat, if you want to
// see area in rgb preview, change mGray to mRgba
/*Imgproc.rectangle(mRgba, eyearea_left.tl(), eyearea_left.br(),
new Scalar(255, 0, 0, 255), 2);
Imgproc.rectangle(mRgba, eyearea_right.tl(), eyearea_right.br(),
new Scalar(255, 0, 0, 255), 2);*/
Core.rectangle(mRgba, eyearea_left.tl(), eyearea_left.br(),
new Scalar(255, 0, 0, 255), 2);
Core.rectangle(mRgba, eyearea_right.tl(), eyearea_right.br(),
new Scalar(255, 0, 0, 255), 2);
if (learn_frames < 5) {
// no learned frames -> Learn templates from at least 5 frames..
templateR = get_template(mJavaDetectorEye, eyearea_right, 24);
templateL = get_template(mJavaDetectorEye, eyearea_left, 24);
learn_frames++;
} else {
// Learning finished, use the new templates for template
// matching
match_eye(eyearea_right, templateR, method);
match_eye(eyearea_left, templateL, method);
}
// cut eye areas and put them to zoom windows
Imgproc.resize(mRgba.submat(eyearea_left), mZoomWindow2,
mZoomWindow2.size());
Imgproc.resize(mRgba.submat(eyearea_right), mZoomWindow,
mZoomWindow.size());
}
return mRgba;
}
private void setMinFaceSize(float faceSize) {
mRelativeFaceSize = faceSize;
mAbsoluteFaceSize = 0;
}
private void CreateAuxiliaryMats() {
if (mGray.empty())
return;
int rows = mGray.rows();
int cols = mGray.cols();
if (mZoomWindow == null) {
mZoomWindow = mRgba.submat(rows / 2 + rows / 10, rows, cols / 2
+ cols / 10, cols);
mZoomWindow2 = mRgba.submat(0, rows / 2 - rows / 10, cols / 2
+ cols / 10, cols);
}
}
private void match_eye(Rect area, Mat mTemplate, int type) {
Point matchLoc;
Mat mROI = mGray.submat(area);
int result_cols = mROI.cols() - mTemplate.cols() + 1;
int result_rows = mROI.rows() - mTemplate.rows() + 1;
// Check for bad template size
if (mTemplate.cols() == 0 || mTemplate.rows() == 0) {
return ;
}
Mat mResult = new Mat(result_cols, result_rows, CvType.CV_8U);
switch (type) {
case TM_SQDIFF:
Imgproc.matchTemplate(mROI, mTemplate, mResult, Imgproc.TM_SQDIFF);
break;
case TM_SQDIFF_NORMED:
Imgproc.matchTemplate(mROI, mTemplate, mResult,
Imgproc.TM_SQDIFF_NORMED);
break;
case TM_CCOEFF:
Imgproc.matchTemplate(mROI, mTemplate, mResult, Imgproc.TM_CCOEFF);
break;
case TM_CCOEFF_NORMED:
Imgproc.matchTemplate(mROI, mTemplate, mResult,
Imgproc.TM_CCOEFF_NORMED);
break;
case TM_CCORR:
Imgproc.matchTemplate(mROI, mTemplate, mResult, Imgproc.TM_CCORR);
break;
case TM_CCORR_NORMED:
Imgproc.matchTemplate(mROI, mTemplate, mResult,
Imgproc.TM_CCORR_NORMED);
break;
}
Core.MinMaxLocResult mmres = Core.minMaxLoc(mResult);
// there is difference in matching methods - best match is max/min value
if (type == TM_SQDIFF || type == TM_SQDIFF_NORMED) {
matchLoc = mmres.minLoc;
} else {
matchLoc = mmres.maxLoc;
}
Point matchLoc_tx = new Point(matchLoc.x + area.x, matchLoc.y + area.y);
Point matchLoc_ty = new Point(matchLoc.x + mTemplate.cols() + area.x,
matchLoc.y + mTemplate.rows() + area.y);
/*Imgproc.rectangle(mRgba, matchLoc_tx, matchLoc_ty, new Scalar(255, 255, 0,
255));*/
Core.rectangle(mRgba, matchLoc_tx, matchLoc_ty, new Scalar(255, 255, 0,
255));
Rect rec = new Rect(matchLoc_tx,matchLoc_ty);
}
private Mat get_template(CascadeClassifier clasificator, Rect area, int size) {
Mat template = new Mat(); //prepare a Mat which will serve as a template for eyes.
Mat mROI = mGray.submat(area); //detect only region of interest which is represented by the area. So, from the total Mat get only the submat that represent roi.
MatOfRect eyes = new MatOfRect(); //will be around eyes (including eyes), this will be filled by detectMultiScale
Point iris = new Point(); //to identify iris.
Rect eye_template = new Rect();
clasificator.detectMultiScale(mROI, eyes, 1.15, 2,
Objdetect.CASCADE_FIND_BIGGEST_OBJECT
| Objdetect.CASCADE_SCALE_IMAGE, new Size(30, 30),
new Size());
Rect[] eyesArray = eyes.toArray(); //get the detected eyes
for (int i = 0; i < eyesArray.length;) {
Rect e = eyesArray[i];
e.x = area.x + e.x; //the starting x coordinates of the rect (area) around the eye + the area
e.y = area.y + e.y;
Rect eye_only_rectangle = new Rect((int) e.tl().x,
(int) (e.tl().y + e.height * 0.4), (int) e.width,
(int) (e.height * 0.6));
mROI = mGray.submat(eye_only_rectangle);
Mat vyrez = mRgba.submat(eye_only_rectangle);
Core.MinMaxLocResult mmG = Core.minMaxLoc(mROI);
// Imgproc.circle(vyrez, mmG.minLoc, 2, new Scalar(255, 255, 255, 255), 2);
Core.circle(vyrez, mmG.minLoc, 2, new Scalar(255, 255, 255, 255), 2);
iris.x = mmG.minLoc.x + eye_only_rectangle.x;
iris.y = mmG.minLoc.y + eye_only_rectangle.y;
eye_template = new Rect((int) iris.x - size / 2, (int) iris.y
- size / 2, size, size);
/*Imgproc.rectangle(mRgba, eye_template.tl(), eye_template.br(),
new Scalar(255, 0, 0, 255), 2);*/
Core.rectangle(mRgba, eye_template.tl(), eye_template.br(),
new Scalar(255, 0, 0, 255), 2);
template = (mGray.submat(eye_template)).clone();
return template;
}
return template;
}
public void onRecreateClick(View v)
{
learn_frames = 0;
}
}
请注意，摄像机已成功打开以进行记录，并且已加载xml文件，但此后没有任何反应.出于测试目的，我将窗口大小设置为100 x 100，我知道它应该是1 x 1.
Notice that the camera opens successfully for recording, and the xml files are loaded, but nothing happens after that. I made the window size as 100 x 100 just for testing purposes, I know it should be 1 x 1.
谁能告诉我如何解决这个问题?如何从后台服务运行opencv摄像机进行面部和眼睛跟踪?
Can anyone please tell me how to solve this problem? How can I run opencv video camera for face and eye tracking from background service?
推荐答案
我试图像在做的那样在服务中获取opencv摄像头，但是我无法同时获取onCameraFrame和onCameraViewStarted回调，这意味着该摄像头没有初始化.经过一堆尝试:
I tried to get the opencv camera in a service as you are doing but I was unable to get neither onCameraFrame nor onCameraViewStarted callbacks, which meant that the camera was not getting initialized. After a bunch of tries:
- 将预览设置为不可见/已消失"->不起作用
- 将预览尺寸设置为1×1或像素大小 相机的宽高比4x3->不起作用
- 在屏幕外设置预览->不起作用
- Setting the preview to INVISIBLE/GONE -> not working
- Setting the preview size to a pixel size of 1×1 or respecting camera's aspect ratio 4x3 ->not working
- Setting the preview outside the screen -> not working
我发现需要用视图的大小预览opencv相机，只有这样我才能够获取onCameraFrame回调. 幸运的是，我可以在摄像头预览上方放置另一个元素以将其隐藏，并仅显示警报.
I found out that opencv camera needs to be previewed with view's size, only that way I was able to get onCameraFrame callback. Fortunately, I could place another element on top of the camera preview to hide it, and show the alarms only.
您可以在此处找到一个简单的CameraInService示例，希望它有用为你.
You could find a simple CameraInService example here, hope it is useful for you.
这篇关于在Android服务中运行OpenCV眼睛检测的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
查看全文
