一、前言
由于公司要求需要提供一个人脸检测的桌面软件打包出去用,目前虹软在这块儿算是做的比较的好的,并且人脸识别也是免费,于是首当其冲的选择了虹软的产品.没相当用javaFX打包之后遇到了各种坑,心累呀.
二、踩坑锦集
-
多线程使用虹软库的时候,必须每个线程一个虹软的引擎,不然会报错.如果想使用线程池,首先继承thread,在对象中封装虹软的引擎对象(这里java使用的是JNA),然后创建线程池的时候,new的thread对象一定是自己创建的那个对象,这样就保证了一个线程一个虹软的对象.详情看一下代码:
/** 封装FD */ public class Detection { private static final int FD_WORK_BUF_SIZE = 20 * 1024 * 1024;// 20 MB private static final int nScale = 16; // 有效值范围[2,50] private static final int MAX_FACE_NUM = 16; // 有效值范围[1,50] private Pointer hFDEngine; private Pointer pFDWorkMem; private boolean doing = false; //在构造方法中,初始化虹软引擎 public Detection(){ String APPID = null; String FD_SDKKEY = null; if(OSUtils.getCurrentOSType() == OSUtils.OSType.Windows){ APPID = ArcFace.appid_Windows_x64; FD_SDKKEY = ArcFace.sdkkey_FD_Windows_x64; } else if(OSUtils.getCurrentOSType() == OSUtils.OSType.Linux){ APPID = ArcFace.appid_Linux_x64; FD_SDKKEY = ArcFace.sdkkey_FD_Linux_x64; } pFDWorkMem = CLibrary.INSTANCE.malloc(FD_WORK_BUF_SIZE); PointerByReference phFDEngine = new PointerByReference(); NativeLong ret = AFD_FSDKEngine.INSTANCE.AFD_FSDK_InitialFaceEngine( APPID, FD_SDKKEY, pFDWorkMem, FD_WORK_BUF_SIZE, phFDEngine, FSDK_OrientPriority.FSDK_OPF_0_HIGHER_EXT, nScale, MAX_FACE_NUM); if (ret.longValue() != 0) { CLibrary.INSTANCE.free(pFDWorkMem); System.err.println(String.format("AFD_FSDK_InitialFaceEngine ret 0x%x %s", ret.longValue(), Error.getErrorMsg(ret.longValue()))); } hFDEngine = phFDEngine.getValue(); } public FaceInfo[] faceDetection(ASVLOFFSCREEN inputImg) { FaceInfo[] faceInfo = new FaceInfo[0]; PointerByReference ppFaceRes = new PointerByReference(); doing = true; NativeLong ret = AFD_FSDKEngine.INSTANCE.AFD_FSDK_StillImageFaceDetection( hFDEngine, inputImg, ppFaceRes); doing = false; if (ret.longValue() != 0) { System.out.println(String.format("AFD_FSDK_StillImageFaceDetection ret 0x%x %s", ret.longValue(), Error.getErrorMsg(ret.longValue()))); return faceInfo; } AFD_FSDK_FACERES faceRes = new AFD_FSDK_FACERES(ppFaceRes.getValue()); if (faceRes.nFace > 0) { faceInfo = new FaceInfo[faceRes.nFace]; for (int i = 0; i < faceRes.nFace; i++) { MRECT rect = new MRECT(new Pointer(Pointer.nativeValue(faceRes.rcFace.getPointer()) + faceRes.rcFace.size() * i)); int orient = faceRes.lfaceOrient.getPointer().getInt(i * 4); faceInfo[i] = new FaceInfo(); faceInfo[i].left = rect.left; faceInfo[i].top = rect.top; faceInfo[i].right = rect.right; faceInfo[i].bottom = rect.bottom; faceInfo[i].orient = orient; } } return faceInfo; } public FaceInfo[] faceDetection(BufferedImage img){ return faceDetection(ArcFaceUtil.getASVLOFFSCREEN(img)); } public void close(){ try { while (doing){ Thread.sleep(200); } AFD_FSDKEngine.INSTANCE.AFD_FSDK_UninitialFaceEngine(hFDEngine); CLibrary.INSTANCE.free(pFDWorkMem); }catch (Exception e){ e.printStackTrace(); } } } //继承线程对象 public class FaceThread extends Thread { private final Detection detection; public FaceThread(Runnable runnable){ super(runnable); detection = new Detection(); } public Detection getDetection() { return detection; } } //创建线程池 public static final ListeningExecutorService FACE_THREAD_POOL = MoreExecutors.listeningDecorator(Executors.newFixedThreadPool(4,new ThreadFactoryBuilder().setThreadFactory(FaceThread::new).setNameFormat("face-%d-Thread").setUncaughtExceptionHandler((t, e) -> e.printStackTrace(System.err)).setDaemon(true).build()));
使用的时候,只需要拿到当前线程对象,就可以了,因为当前线程对象里面有虹软的识别引擎对象.
-
某些电脑加载不了虹软的dll文件
这个首先排查自己代码中的dll路径有没有错,如果没有错,就看库的版本与jvm的版本是否一致,必须两个都为32或者64才能跑.妈的如果这时候版本也是一致的,就看电脑上是否安装了vc++ 2013的运行库,如果没有就下载一个进行安装.如果还是没办法,问官方吧.
-
打包成jar加载不了dll文件
这个的确是个问题,对于一个有强迫症的人来说.因为dll文件被打包进jar中了,所以dll就不存在文件系统中了,这个时候需要从jar中释放出dll文件来,在进行加载dll.代码如下:
/** * @author * */ public class LoadUtils { public static <T> T loadOSLibrary(String libName, Class<T> interfaceClass) { try { String filePath = ""; URI uri = LoadUtils.class.getProtectionDomain().getCodeSource().getLocation().toURI(); if (Platform.isWindows()) { filePath = "arcsoft/windows_x64/" + libName + ".dll"; uri = uri.resolve(filePath); } else if (Platform.isLinux()) { filePath = "arcsoft/linux_x64/" + libName + ".so"; uri = uri.resolve(filePath); } else { System.out.println("unsupported platform"); System.exit(0); } Path path = Paths.get(uri); //尝试释放资源 if(!Files.exists(path) && filePath.length() > 0){ Path parent = Paths.get(uri.resolve("./")); System.out.println(parent); //父目录不存在,创建目录 if(!Files.exists(parent)){ Files.createDirectories(parent); } Files.copy(LoadUtils.class.getResourceAsStream("/" + filePath),path); } System.out.println(path.toString()); return loadLibrary(path.toString(), interfaceClass); } catch (URISyntaxException | IOException e) { e.printStackTrace(); } return null; } private static <T> T loadLibrary(String filePath, Class<T> interfaceClass) { return Native.loadLibrary(filePath, interfaceClass); } }
-
使用虹软的ft(libarcsoft_fsdk_face_tracking.dll)和fd(libarcsoft_fsdk_face_detection.dll)的效果是不一样的.
最开始使用的ft进行的人脸检测,发现有时候能检测出人脸而有时候却检测不出人脸,坑啊.最后换了fd之后这种情况就没了.
三、总结
项目不止,踩坑不断,后面踩到新坑,会补充上来.心累……