客服电话
APP下载
迪恩网络APP
随时随地掌握行业动态
官方微信
扫描二维码
关注迪恩网络微信公众号
|
手上有一个项目,需要检验使用本程序的,是否本人!因为在程序使用前,我们都已经做过头像现场采集,所以源头呢是不成问题的,那么人脸检测,人脸比对,怎么办呢?度娘了下,目前流行的几个人脸检测,人脸比对核心,大多都是基于互联网的,但我们的项目是基于本地服务器,那就有点麻烦了,后来找到ArcFace.它的核心允许本地调用,那就好办了,立刻去了他们的网站下载sdk,看论坛,下DEMO;我当时下的是这个:ArcFace C#DEMO public delegate bool MatchHandler(string userid, string studyid, string photoid, string photopath); //最后我的Program里边,就是做一个递归,去不断的问数据库拿被标志需要进行核对的记录,拿到图片后,就进行比对; MatchHandler handler = new MatchHandler(MatchUserFace.GetAndMatchImage); string Identification = string.Format("USERID:{0} STUDYID:{1} PHID:{2}", userid, studyid, photoid); IAsyncResult result = handler.BeginInvoke(userid, studyid, photoid, pathstr, RecognizeEngine, DetectEngine, new AsyncCallback(CallbackFunc), Identification); 下边这段就是异步的结果回调; static void CallbackFunc(IAsyncResult result) { MatchHandler handler = (MatchHandler)((AsyncResult)result).AsyncDelegate; bool match = handler.EndInvoke(result); string strmatch = string.Empty; if (match) { strmatch = " 比对结果:OK"; } else { strmatch = " 比对结果:NO"; } Console.WriteLine(result.AsyncState + strmatch); GC.Collect(); } 写好了,发布到服务器上,还想着中午吃个鸡腿奖励下自己;不想…发布后不到两小时,小弟来说:服务器是不是出问题了,下边所有业务窗口访问速度严重延迟…立马跑到机房去看,一看没毛病呀,所有的服务都好好的,没有卦死…再打开资源监视器一看,靠…那个比对端一下吃3个多G的内存,而且还在不断上升中…立马停掉,然后再问小弟,下边业务是否正常,他回复正常了…那么说,就是我写的这个比对端有问题了!改!!! public delegate bool MatchHandler(string userid, string studyid, string photoid, string photopath, IntPtr RecognizeEngine, IntPtr DetectEngine); //然后初始化SDK放到了Program里做: string appId = "4yHjnxK94FCK6L7HaJieWawSLubnANXXXXX"; string sdkFDKey = "7S6Xp4mtroLnjTt7qDYnd2dqHXXXXX"; string sdkFRKey = "7S6Xp4mtroLnjTt7qDYnd2dxSgXXXXX"; int retCode = AFDFunction.AFD_FSDK_InitialFaceEngine(appId, sdkFDKey, pMem, detectSize, ref DetectEngine, 5, nScale, nMaxFaceNum); int retCode2 = AFRFunction.AFR_FSDK_InitialEngine(appId, sdkFRKey, pMemRecongnize, detectSize, ref RecognizeEngine); //最后把异步调用的方法改成如下: MatchHandler handler = new MatchHandler(MatchUserFace.GetAndMatchImage); string Identification = string.Format("USERID:{0} STUDYID:{1} PHID:{2}", userid, studyid, photoid); IAsyncResult result = handler.BeginInvoke(userid, studyid, photoid, pathstr, RecognizeEngine, DetectEngine, new AsyncCallback(CallbackFunc), Identification); 再次发布到服务器.然后再到资源监视器去看,哟…线程数不高了而且增长的还不快…好开心!!以为搞好了;就回宿舍睡觉去了!!不想…睡得迷糊的时候,我们的客服小妹妹的电话就打到我这了,我说什么事,她说现在大面积反映用户比对不了?what?我说不可能吧,是不是当地电信故障呀?我自己拿手机试了下,真的不行呀!!!快速赶回办公室远程看了下服务器,我的乖乖…比对端卦了!!!我再看日志,日志没有捕捉到程序异常,只是捕到了个:Value cannot be null.Parameter name: source;我吃你大米了,我刨你家玉米地了,为啥要这么对我!重启比对端,然后都可以正常运作了…我决定在这监视这个比对端,在资源监视器我到是发现了一个:w3wp.exe它在不断的涨内存(这是要划重点的)想想这可已经是深夜了.果不出其然,运行了大概两个多小时后,程序又卦了.我的乖乖,为啥会这样呢,一时半会也想不出办法呀!我也总不能呆在服务器旁它停了,我就重启吧! private static byte[] detectAndExtractFeature(Image imageParam, out Image facerect, IntPtr RecognizeEngine, IntPtr DetectEngine) { byte[] feature = null; facerect = null; try { int width = 0; int height = 0; int pitch = 0; Bitmap bitmap = new Bitmap(imageParam); byte[] imageData = getBGR(bitmap, ref width, ref height, ref pitch); IntPtr imageDataPtr = Marshal.AllocHGlobal(imageData.Length); Marshal.Copy(imageData, 0, imageDataPtr, imageData.Length); ASVLOFFSCREEN offInput = new ASVLOFFSCREEN(); offInput.u32PixelArrayFormat = 513; offInput.ppu8Plane = new IntPtr[4]; offInput.ppu8Plane[0] = imageDataPtr; offInput.i32Width = width; offInput.i32Height = height; offInput.pi32Pitch = new int[4]; offInput.pi32Pitch[0] = pitch; AFD_FSDK_FACERES faceRes = new AFD_FSDK_FACERES(); IntPtr offInputPtr = Marshal.AllocHGlobal(Marshal.SizeOf(offInput)); Marshal.StructureToPtr(offInput, offInputPtr, false); IntPtr faceResPtr = Marshal.AllocHGlobal(Marshal.SizeOf(faceRes)); //人脸检测 int detectResult = AFDFunction.AFD_FSDK_StillImageFaceDetection(DetectEngine, offInputPtr, ref faceResPtr); if (detectResult == 0) { try { object obj = Marshal.PtrToStructure(faceResPtr, typeof(AFD_FSDK_FACERES)); faceRes = (AFD_FSDK_FACERES)obj; for (int i = 0; i < faceRes.nFace; i++) { MRECT rect = (MRECT)Marshal.PtrToStructure(faceRes.rcFace + Marshal.SizeOf(typeof(MRECT)) * i, typeof(MRECT)); int orient = (int)Marshal.PtrToStructure(faceRes.lfaceOrient + Marshal.SizeOf(typeof(int)) * i, typeof(int)); if (i == 0) { facerect = CutFace(bitmap, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top); } } } catch (Exception ex) { LogNetWriter.Error("人脸检测时出错:" + ex.Message); } } if (faceRes.nFace > 0) { try { AFR_FSDK_FaceInput faceResult = new AFR_FSDK_FaceInput(); int orient = (int)Marshal.PtrToStructure(faceRes.lfaceOrient, typeof(int)); faceResult.lOrient = orient; faceResult.rcFace = new MRECT(); MRECT rect = (MRECT)Marshal.PtrToStructure(faceRes.rcFace, typeof(MRECT)); faceResult.rcFace = rect; IntPtr faceResultPtr = Marshal.AllocHGlobal(Marshal.SizeOf(faceResult)); Marshal.StructureToPtr(faceResult, faceResultPtr, false); AFR_FSDK_FaceModel localFaceModels = new AFR_FSDK_FaceModel(); IntPtr localFaceModelsPtr = Marshal.AllocHGlobal(Marshal.SizeOf(localFaceModels)); int extractResult = AFRFunction.AFR_FSDK_ExtractFRFeature(RecognizeEngine, offInputPtr, faceResultPtr, localFaceModelsPtr); if (extractResult == 0) { Marshal.FreeHGlobal(faceResultPtr); Marshal.FreeHGlobal(offInputPtr); object objFeature = Marshal.PtrToStructure(localFaceModelsPtr, typeof(AFR_FSDK_FaceModel)); Marshal.FreeHGlobal(localFaceModelsPtr); localFaceModels = (AFR_FSDK_FaceModel)objFeature; feature = new byte[localFaceModels.lFeatureSize]; Marshal.Copy(localFaceModels.pbFeature, feature, 0, localFaceModels.lFeatureSize); localFaceModels = new AFR_FSDK_FaceModel(); } } catch (Exception ex) { LogNetWriter.Error("提取特征时出错:" + ex.Message); } } bitmap.Dispose(); imageData = null; Marshal.FreeHGlobal(imageDataPtr); //Marshal.FreeHGlobal(faceResPtr); offInput = new ASVLOFFSCREEN(); faceRes = new AFD_FSDK_FACERES(); } catch (Exception ex) { LogNetWriter.Error("识别人脸并提取人脸特征出错:" + ex.Message); } return feature; } 当然了,比对的时候也作了一些修改,就是当比对完了以后,就做了指针释放; Marshal.FreeHGlobal(firstFeaturePtr);
Marshal.FreeHGlobal(secondFeaturePtr);
Marshal.FreeHGlobal(firstPtr);
Marshal.FreeHGlobal(secondPtr);
经过这一次修改后,再发布到服务器,哟…不错哦…运行的时间久了…但还是会卦,而且那个w3wp.exe还是会不断的拉内存;这个版本的运行时间可以达到4小左右了;我就想总得有个解决办法吧;再次致电虹软,再次反映这个问题,虹软这边给我的建议就是不要去进行多线程,我想想也对,要把逻辑简单化,我就把识别核心打包成一个EXE.然后在Program里调用这个EXE.意思就是每当我有需要识别的图片,我就调一个EXE.然后EXE处理完以后,就自我释放了… //这里就是一条线程在做处理 string strmatch = string.Empty; ControlExeClass _ControlExeClass = new Model.ControlExeClass(); //这个方法是调一个EXE,EXE的内容是:ControlExeClass.cs; //做的任务就是把图片进行人脸检测,人脸特征提取,人脸识别; bool bo = _ControlExeClass.ControlExe(userid, studyid, photoid, pathstr); if (bo) { iCheck_OK++; label5.Text = iCheck_OK.ToString(); strmatch = " 比对结果:OK"; } else { strmatch = " 比对结果:NO"; } string dates = " 比对时间:" + DateTime.Now.ToString("yyyy-MM-dd HH:mm:ss"); textBox1.Text += " USERID:" + userid + " STUDYID:" + studyid + " PHID:" + photoid + strmatch + dates + Environment.NewLine; 然后那个EXE就是沿用MatchUserFace调用类,在EXE的主线程里完成调用;还别说,用了这个方法后,内存不拉升了,而且w3wp.exe上涨,也只是在EXE工作的一刹那上来,EXE干完活后,它就会生成一个新的w3wp.exe,旧的w3wp.exe那个会被注销掉…哗…想想就开心,终于如愿解决了问题,但…当一个人觉得越顺利时,往往大麻烦就会来了.正如我觉得上天不会对我那么好一样,运行了大概一天后,程序还是卦了.苍天呀,大地呀,我到底做错了什么… private static void CycleData() { while (true) { if (_DoWork) { break; } else { QueryDataFile("U"); Thread.Sleep(1500); } Thread.Sleep(2000); } } 至此所有问题解决!!!哦…忘说了,我后来没有单独调用EXE这种方法了,改成了第一版的控制台程序,其结果是一样的; |
请发表评论