安卓人臉檢測優化
一、人臉檢測模組移植
1.拷貝opencv-3.3.0-android-sdk\OpenCV-android-sdk\samples\face-detection\jni目錄到工程app module的main目錄下
2.修改jni目錄下的Android.mk
(1) 將
#OPENCV_INSTALL_MODULES:=off
#OPENCV_LIB_TYPE:=SHARED
修改為:
OPENCV_INSTALL_MODULES:=on
OPENCV_LIB_TYPE:=SHARED
其中,OPENCV_INSTALL_MODULES的作用是在打包apk時載入OpenCV的動態庫;OPENCV_LIB_TYPE的作用是指定OpenCV庫的型別為動態庫。
(2) 將
ifdef OPENCV_ANDROID_SDK
ifneq ("","$(wildcard $(OPENCV_ANDROID_SDK)/OpenCV.mk)")
include ${OPENCV_ANDROID_SDK}/OpenCV.mk
else
include ${OPENCV_ANDROID_SDK}/sdk/native/jni/OpenCV.mk
endif
include ../../sdk/native/jni/OpenCV.mk
endif
修改為:
include E:\Environment\opencv-3.3.0-android-sdk\OpenCV-android-sdk\sdk\native\jni\OpenCV.mk
其中,include包含的就是OpenCV SDK中OpenCV.mk檔案所儲存的絕對路徑。最終Android.mk修改效果如下:
3.修改jni目錄下Application.mk。由於在匯入OpenCV libs時只拷貝了armeabi 、armeabi-v7a、arm64-v8a,因此這裡指定編譯平臺也為上述三個;修改APP_PLaTFORM版本為android-16(可根據自身情況而定),具體如下:
APP_STL := gnustl_static
APP_CPPFLAGS := -frtti –fexceptions
# 指定編譯平臺
APP_ABI := armeabi armeabi-v7a arm64-v8a
# 指定Android平臺
APP_PLATFORM := android-16
4.修改DetectionBasedTracker_jni.h和DetectionBasedTracker_jni.cpp檔案,將原始檔中所有包含字首“Java_org_opencv_samples_facedetect_”替換為“Java_com_jiangdg_opencv4android_natives_”,其中com.jiangdg.opencv4android.natives是Java層類DetectionBasedTracker.java所在的包路徑,該類包含了人臉檢測相關的native方法,否則,在呼叫自己編譯生成的so庫時會提示找不到該本地函式錯誤,以DetectionBasedTracker_jni.h為例:
/* DO NOT EDIT THIS FILE - it is machine generated */
#include <jni.h>
/* Header for class org_opencv_samples_fd_DetectionBasedTracker */
#ifndef _Included_org_opencv_samples_fd_DetectionBasedTracker
#define _Included_org_opencv_samples_fd_DetectionBasedTracker
#ifdef __cplusplus
extern "C" {
#endif
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeCreateObject
* Signature: (Ljava/lang/String;F)J
*/
JNIEXPORT jlong JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeCreateObject
(JNIEnv *, jclass, jstring, jint);
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeDestroyObject
* Signature: (J)V
*/
JNIEXPORT void JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeDestroyObject
(JNIEnv *, jclass, jlong);
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeStart
* Signature: (J)V
*/
JNIEXPORT void JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeStart
(JNIEnv *, jclass, jlong);
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeStop
* Signature: (J)V
*/
JNIEXPORT void JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeStop
(JNIEnv *, jclass, jlong);
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeSetFaceSize
* Signature: (JI)V
*/
JNIEXPORT void JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeSetFaceSize
(JNIEnv *, jclass, jlong, jint);
/*
* Class: org_opencv_samples_fd_DetectionBasedTracker
* Method: nativeDetect
* Signature: (JJJ)V
*/
JNIEXPORT void JNICALL Java_com_jiangdg_opencv4android_natives_DetectionBasedTracker_nativeDetect
(JNIEnv *, jclass, jlong, jlong, jlong);
#ifdef __cplusplus
}
#endif
#endif
5.開啟Android Studio中的Terminal視窗,使用cd命令切換到工程jni目錄所在位置,並執行ndk-build命令,然後會自動在工程的app/src/main目錄下生成libs和obj目錄,其中libs目錄存放的是目標動態庫libdetection_based_tracker.so。
生成so庫:
注意:如果執行ndk-build命令提示命令不存在,說明你的ndk環境變數沒有配置好。
6.修改app模組build.gradle中的sourceSets欄位,禁止自動呼叫ndk-build命令,設定目標so的存放路徑,程式碼如下:
android {
compileSdkVersion 25
defaultConfig {
applicationId "com.jiangdg.opencv4android"
minSdkVersion 15
targetSdkVersion 25
versionCode 1
versionName "1.0"
}
….// 程式碼省略
sourceSets {
main {
jni.srcDirs = [] //禁止自動呼叫ndk-build命令
jniLibs.srcDir 'src/main/libs' // 設定目標的so存放路徑
}
}
….// 程式碼省略
}
其中,jni.srcDirs = []的作用是禁用gradle預設的ndk-build,防止AS自己生成android.mk編譯jni工程,jniLibs.srcDir 'src/main/libs'的作用設定目標的so存放路徑,以將自己生成的so組裝到apk中。
二、原始碼解析
使用OpenCV3.3.0庫實現人臉檢測功能主要包含以下四個步驟,即:
(1) 初始化載入OpenCV庫引擎;
(2) 通過OpenCV庫開啟Camera渲染;
(3) 載入人臉檢測模型;
(4) 呼叫人臉檢測本地動態庫實現人臉識別;
1.初始化載入OpenCV庫引擎
OpenCV庫的載入有兩種方式,一種通過OpenCV Manager進行動態載入,也就是官方推薦的方式,這種方式需要另外安裝OpenCV Manager,主要通過呼叫OpenCVLoader.initAsync()方法進行初始化;另一種為靜態載入,也就是本文所使用的方法,即先將相關架構的so包拷貝到工程的libs目錄,通過呼叫OpenCVLoader.initDebug()方法進行初始化,類似於呼叫system.loadLibrary("opencv_java")。
if (!OpenCVLoader.initDebug()) {
// 靜態載入OpenCV失敗,使用OpenCV Manager初始化
// 引數:OpenCV版本;上下文;載入結果回撥介面
OpenCVLoader.initAsync(OpenCVLoader.OPENCV_VERSION_3_3_0,
this, mLoaderCallback);
} else {
// 如果靜態載入成功,直接呼叫onManagerConnected方法
mLoaderCallback.onManagerConnected(LoaderCallbackInterface.SUCCESS);
}
其中,mLoaderCallback為OpenCV庫初始化狀態回撥介面,當OpenCV被初始化成功後其onManagerConnected(int status)方法會被呼叫,而我們就可以在該方法中處理本地動態庫的載入、載入人臉檢測模型檔案、初始化人臉檢測引擎以及開啟Camera渲染等操作,具體程式碼如下:
private BaseLoaderCallback mLoaderCallback = new BaseLoaderCallback(this) {
@Override
public void onManagerConnected(int status) {
switch (status) {
case LoaderCallbackInterface.SUCCESS:
// OpenCV初始化載入成功,再載入本地so庫
System.loadLibrary("detection_based_tracker");
// 載入人臉檢測模型
…..
// 初始化人臉檢測引擎
…..
// 開啟渲染Camera
mCameraView.enableView();
break;
default:
super.onManagerConnected(status);
break;
}
}
};
2. 通過OpenCV庫開啟Camera渲染
在OpenCV中與Camera緊密相關的主要有兩個類,即CameraBridgeViewBase和JavaCameraView,其中,CameraBridgeViewBase是一個基類,繼承於SuarfaceView和SurafaceHolder.Callback介面,用於實現Camera與OpenCV庫之間的互動,它主要的作用是控制Camera、處理視訊幀以及呼叫相關內部介面對視訊幀做相關調整,然後將調整後的視訊幀資料渲染到手機螢幕上。比如enableView()方法、disableView()方法用於連線到Camera和斷開與Camera的連線,程式碼如下:
public void enableView() {
synchronized(mSyncObject) {
mEnabled = true;
checkCurrentState();
}
}
public void disableView() {
synchronized(mSyncObject) {
mEnabled = false;
checkCurrentState();
}
}
其中,checkCurrentState()方法用於更新Camera的渲染狀態,它呼叫了processEnterState()方法來啟動或停用Camera,以及將Camera的狀態對外回撥。為了方便開發者實時獲取Camera的連線狀態,CameraBridgeViewBase還提供了一個setCvCameraViewListener(CvCameraViewListener2 listener)方法,引數listener其一個內部介面,它包括三個方法:onCameraViewStarted(int width, int height)、void onCameraViewStopped()、Mat onCameraFrame(CvCameraViewFrame inputFrame),分別用於對外回撥Camera連線狀態和傳遞Camera的實時視訊幀資料。
private void checkCurrentState() {
Log.d(TAG, "call checkCurrentState");
int targetState;
if (mEnabled && mSurfaceExist && getVisibility() == VISIBLE) {
targetState = STARTED;
} else {
targetState = STOPPED;
}
if (targetState != mState) {
/* The state change detected. Need to exit the current state and enter target state */
processExitState(mState);
mState = targetState;
processEnterState(mState);
}
}
private void processEnterState(int state) {
Log.d(TAG, "call processEnterState: " + state);
switch(state) {
case STARTED:
// 呼叫connectCamera()抽象方法,啟動Camera
onEnterStartedState();
// 呼叫連線成功監聽器介面方法
if (mListener != null) {
mListener.onCameraViewStarted(mFrameWidth, mFrameHeight);
}
break;
case STOPPED:
// 呼叫disconnectCamera()抽象方法,停用Camera
onEnterStoppedState();
// 呼叫斷開連線監聽器介面方法
if (mListener != null) {
mListener.onCameraViewStopped();
}
break;
};
}
既然CameraBridgeViewBase是一個基類,與Camera緊密相關的connectCamera()和disconnectCamera()又是抽象方法,那麼就必定會有一個子類來實現這兩個方法,而這個子類就是JavaCameraView。JavaCameraView繼承於CameraBridgeViewBase和PreviewCallback介面,是銜接OpenCV和Camera的橋樑,是Camera啟動、禁止的實際實現者,在這個類裡我們可以看到關於Camera很多熟悉的操作。原始碼如下:
@Override
protected boolean connectCamera(int width, int height) {
// 初始化Camera,連線到Camera
if (!initializeCamera(width, height))
return false;
mCameraFrameReady = false;
// 開啟一個與Camera相關的工作執行緒CameraWorker
Log.d(TAG, "Starting processing thread");
mStopThread = false;
mThread = new Thread(new CameraWorker());
mThread.start();
return true;
}
@Override
protected void disconnectCamera() {
// 斷開Camera連線,釋放相關資源
try {
mStopThread = true;
Log.d(TAG, "Notify thread");
synchronized (this) {
this.notify();
}
// 停止工作執行緒
if (mThread != null)
mThread.join();
} catch (InterruptedException e) {
e.printStackTrace();
} finally {
mThread = null;
}
/* Now release camera */
releaseCamera();
mCameraFrameReady = false;
}
CameraWorker是一個工作執行緒,用於處理從onPreviewFrame獲得的視訊幀資料,其儲存在一個Mat型別的陣列中。它會不斷呼叫父類CameraBridgeViewBase的deliverAndDrawFrame方法,將處理後的視訊幀資料流通過呼叫內部介面CvCameraViewListener2的onCameraFrame(CvCameraViewFrame frame)對外回撥。
private class CameraWorker implements Runnable {
@Override
public void run() {
do {
…..//程式碼省略
if (!mStopThread && hasFrame) {
if (!mFrameChain[1 - mChainIdx].empty())
deliverAndDrawFrame(mCameraFrame[1 - mChainIdx]);
}
} while (!mStopThread);
}
}
3. 載入人臉檢測模型
為了得到更好的人臉檢測效能,OpenCV在SDK中提供了多個frontface檢測器(人臉模型),存放在..\opencv-3.3.0-android-sdk\OpenCV-android-sdk\sdk\etc\目錄下,這篇對OpenCV自帶的人臉檢測模型做了比較,結果顯示LBP實時性要好些。因此,本文選用目lbpcascades錄下lbpcascade_frontalface.xml模型,該模型包括了3000個正樣本和1500個負樣本,我們將其拷貝到AS工程的res/raw目錄下,並通過getDir方法儲存到/data/data/com.jiangdg.opencv4android/ cascade目錄下。
InputStream is = getResources().openRawResource(R.raw.lbpcascade_frontalface);
File cascadeDir = getDir("cascade", Context.MODE_PRIVATE);
mCascadeFile = new File(cascadeDir, "lbpcascade_frontalface.xml");
FileOutputStream os = new FileOutputStream(mCascadeFile);
byte[] buffer = new byte[4096];
int byteesRead;
while ((byteesRead = is.read(buffer)) != -1) {
os.write(buffer, 0, byteesRead);
}
is.close();
os.close();
注:關於模型的訓練在以後的博文中會討論到。
4. 人臉檢測
在opencv-3.3.0-android-sdk的face-detection示例專案中,提供了CascadeClassifier和
DetectionBasedTracker兩種方式來實現人臉檢測,其中,CascadeClassifier是OpenCV用於人臉檢測的一個級聯分類器,DetectionBasedTracker是通過JNI程式設計實現的人臉檢測。兩種方式我都試用了下,發現DetectionBasedTracker方式還是比CascadeClassifier穩定些,CascadeClassifier會存在一定頻率的誤檢。
public class DetectionBasedTracker {
private long mNativeObj = 0;
// 構造方法:初始化人臉檢測引擎
public DetectionBasedTracker(String cascadeName, int minFaceSize) {
mNativeObj = nativeCreateObject(cascadeName, minFaceSize);
}
// 開始人臉檢測
public void start() {
nativeStart(mNativeObj);
}
// 停止人臉檢測
public void stop() {
nativeStop(mNativeObj);
}
// 設定人臉最小尺寸
public void setMinFaceSize(int size) {
nativeSetFaceSize(mNativeObj, size);
}
// 檢測
public void detect(Mat imageGray, MatOfRect faces) {
nativeDetect(mNativeObj, imageGray.getNativeObjAddr(), faces.getNativeObjAddr());
}
// 釋放資源
public void release() {
nativeDestroyObject(mNativeObj);
mNativeObj = 0;
}
// native方法
private static native long nativeCreateObject(String cascadeName, int minFaceSize);
private static native void nativeDestroyObject(long thiz);
private static native void nativeStart(long thiz);
private static native void nativeStop(long thiz);
private static native void nativeSetFaceSize(long thiz, int size);
private static native void nativeDetect(long thiz, long inputImage, long faces);
}
初始化DetectionBasedTracker後,我們只需要在CvCameraViewListener2介面的onCameraFrame方法中對每幀圖片進行人臉檢測即可。
@Override
public Mat onCameraFrame(CameraBridgeViewBase.CvCameraViewFrame inputFrame) {
….// 程式碼省略
// 獲取檢測到的臉部資料
MatOfRect faces = new MatOfRect();
…// 程式碼省略
if (mNativeDetector != null) {
mNativeDetector.detect(mGray, faces);
}
// 繪製檢測框
Rect[] facesArray = faces.toArray();
for (int i = 0; i < facesArray.length; i++) {
Imgproc.rectangle(mRgba, facesArray[i].tl(), facesArray[i].br(), FACE_RECT_COLOR, 3);
}
return mRgba;
}
注:由於篇幅原因,關於人臉檢測的C/C++實現程式碼(原理),我們將在後續文章中討論。
三、效果演示
1. FaceDetectActivity.class
/**
* 人臉檢測
*
* Created by jiangdongguo on 2018/1/4.
*/
public class FaceDetectActivity extends AppCompatActivity implements CameraBridgeViewBase.CvCameraViewListener2 {
private static final int JAVA_DETECTOR = 0;
private static final int NATIVE_DETECTOR = 1;
private static final String TAG = "FaceDetectActivity";
@BindView(R.id.cameraView_face)
CameraBridgeViewBase mCameraView;
private Mat mGray;
private Mat mRgba;
private int mDetectorType = NATIVE_DETECTOR;
private int mAbsoluteFaceSize = 0;
private float mRelativeFaceSize = 0.2f;
private DetectionBasedTracker mNativeDetector;
private CascadeClassifier mJavaDetector;
private static final Scalar FACE_RECT_COLOR = new Scalar(0, 255, 0, 255);
private File mCascadeFile;
private BaseLoaderCallback mLoaderCallback = new BaseLoaderCallback(this) {
@Override
public void onManagerConnected(int status) {
switch (status) {
case LoaderCallbackInterface.SUCCESS:
// OpenCV初始化載入成功,再載入本地so庫
System.loadLibrary("detection_based_tracker");
try {
// 載入人臉檢測模式檔案
InputStream is = getResources().openRawResource(R.raw.lbpcascade_frontalface);
File cascadeDir = getDir("cascade", Context.MODE_PRIVATE);
mCascadeFile = new File(cascadeDir, "lbpcascade_frontalface.xml");
FileOutputStream os = new FileOutputStream(mCascadeFile);
byte[] buffer = new byte[4096];
int byteesRead;
while ((byteesRead = is.read(buffer)) != -1) {
os.write(buffer, 0, byteesRead);
}
is.close();
os.close();
// 使用模型檔案初始化人臉檢測引擎
mJavaDetector = new CascadeClassifier(mCascadeFile.getAbsolutePath());
if (mJavaDetector.empty()) {
Log.e(TAG, "載入cascade classifier失敗");
mJavaDetector = null;
} else {
Log.d(TAG, "Loaded cascade classifier from " + mCascadeFile.getAbsolutePath());
}
mNativeDetector = new DetectionBasedTracker(mCascadeFile.getAbsolutePath(), 0);
cascadeDir.delete();
} catch (FileNotFoundException e) {
e.printStackTrace();
} catch (IOException e) {
e.printStackTrace();
}
// 開啟渲染Camera
mCameraView.enableView();
break;
default:
super.onManagerConnected(status);
break;
}
}
};
@Override
protected void onCreate(@Nullable Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
getWindow().addFlags(WindowManager.LayoutParams.FLAG_FULLSCREEN);
setContentView(R.layout.activity_facedetect);
// 繫結View
ButterKnife.bind(this);
mCameraView.setVisibility(CameraBridgeViewBase.VISIBLE);
// 註冊Camera渲染事件監聽器
mCameraView.setCvCameraViewListener(this);
}
@Override
protected void onResume() {
super.onResume();
// 靜態初始化OpenCV
if (!OpenCVLoader.initDebug()) {
Log.d(TAG, "無法載入OpenCV本地庫,將使用OpenCV Manager初始化");
OpenCVLoader.initAsync(OpenCVLoader.OPENCV_VERSION_3_3_0, this, mLoaderCallback);
} else {
Log.d(TAG, "成功載入OpenCV本地庫");
mLoaderCallback.onManagerConnected(LoaderCallbackInterface.SUCCESS);
}
}
@Override
protected void onPause() {
super.onPause();
// 停止渲染Camera
if (mCameraView != null) {
mCameraView.disableView();
}
}
@Override
protected void onDestroy() {
super.onDestroy();
// 停止渲染Camera
if (mCameraView != null) {
mCameraView.disableView();
}
}
@Override
public void onCameraViewStarted(int width, int height) {
// 灰度影象
mGray = new Mat();
// R、G、B彩色影象
mRgba = new Mat();
}
@Override
public void onCameraViewStopped() {
mGray.release();
mRgba.release();
}
@Override
public Mat onCameraFrame(CameraBridgeViewBase.CvCameraViewFrame inputFrame) {
mRgba = inputFrame.rgba();
mGray = inputFrame.gray();
// 設定臉部大小
if (mAbsoluteFaceSize == 0) {
int height = mGray.rows();
if (Math.round(height * mRelativeFaceSize) > 0) {
mAbsoluteFaceSize = Math.round(height * mRelativeFaceSize);
}
mNativeDetector.setMinFaceSize(mAbsoluteFaceSize);
}
// 獲取檢測到的臉部資料
MatOfRect faces = new MatOfRect();
if (mDetectorType == JAVA_DETECTOR) {
if (mJavaDetector != null) {
mJavaDetector.detectMultiScale(mGray, faces, 1.1, 2, 2,
new Size(mAbsoluteFaceSize, mAbsoluteFaceSize), new Size());
}
} else if (mDetectorType == NATIVE_DETECTOR) {
if (mNativeDetector != null) {
mNativeDetector.detect(mGray, faces);
}
} else {
Log.e(TAG, "Detection method is not selected!");
}
// 繪製檢測框
Rect[] facesArray = faces.toArray();
for (int i = 0; i < facesArray.length; i++) {
Imgproc.rectangle(mRgba, facesArray[i].tl(), facesArray[i].br(), FACE_RECT_COLOR, 3);
}
Log.i(TAG, "共檢測到 " + faces.toArray().length + " 張臉");
return mRgba;
}
}
2. activity_facedetect.xml
<?xml version="1.0" encoding="utf-8"?>
<RelativeLayout xmlns:android="http://schemas.android.com/apk/res/android"
xmlns:opencv="http://schemas.android.com/apk/res-auto"
android:layout_width="match_parent"
android:layout_height="match_parent"
android:orientation="vertical">
<org.opencv.android.JavaCameraView
android:id="@+id/cameraView_face"
android:layout_width="match_parent"
android:layout_height="match_parent"
android:visibility="gone"
opencv:camera_id="any"
opencv:show_fps="true" />
</RelativeLayout>
3. AndroidMnifest.xml
<?xml version="1.0" encoding="utf-8"?>
<manifest xmlns:android="http://schemas.android.com/apk/res/android"
package="com.jiangdg.opencv4android">
<uses-permission android:name="android.permission.CAMERA"/>
<uses-feature android:name="android.hardware.camera" android:required="false"/>
<uses-feature android:name="android.hardware.camera.autofocus" android:required="false"/>
<uses-feature android:name="android.hardware.camera.front" android:required="false"/>
<uses-feature android:name="android.hardware.camera.front.autofocus" android:required="false"/>
<supports-screens android:resizeable="true"
android:smallScreens="true"
android:normalScreens="true"
android:largeScreens="true"
android:anyDensity="true" />
<application
android:allowBackup="true"
android:icon="@mipmap/ic_launcher"
android:label="@string/app_name"
android:roundIcon="@mipmap/ic_launcher_round"
android:supportsRtl="true"
android:theme="@style/AppTheme">
<activity android:name=".MainActivity">
<intent-filter>
<action android:name="android.intent.action.MAIN" />
<category android:name="android.intent.category.LAUNCHER" />
</intent-filter>
</activity>
<activity android:name=".HelloOpenCVActivity"
android:screenOrientation="landscape"
android:configChanges="keyboardHidden|orientation"/>
<activity android:name=".FaceDetectActivity"
android:screenOrientation="landscape"
android:configChanges="keyboardHidden|orientation"/>
</application>
</manifest>