c# 利用AForge和百度AI开发实时人脸识别

baiduAIFaceIdentify项目是C#语言，集成百度AI的SDK利用AForge开发的实时人脸识别的小demo，里边包含了人脸检测识别，人脸注册，人脸登录等功能

人脸实时检测识别功能

思路是利用AForge打开摄像头，通过摄像头获取到的图像显示在winform窗体中AForge的控件中，利用AForge控件中的NewFrame事件获取要显示的每一帧的图像，获取图像传输到百度AI平台进行人脸检测，并且将检测结果反馈到界面显示的图像中。在这个过程中有两个问题，获取图像上传到百度AI平台进行分析需要时间，这个时间跟网络有关，所以需要单独一个线程进行人脸识别，第二个问题，百度人脸识别接口开发者一秒内只能掉用2次接口，所以需要控制不是每一帧的图像都要上传。所以基于以上思路

首先页面初始化的时候获取视频设备、启动一个单独线程控制1秒内人脸检测的次数：

private void Form1_Load(object sender, EventArgs e){/// 获取电脑已经安装的视频设备videoDevices = new FilterInfoCollection(FilterCategory.VideoInputDevice);if (videoDevices!=null && videoDevices.Count>0){foreach (FilterInfo device in videoDevices){comboBox1.Items.Add(device.Name);}comboBox1.SelectedIndex = 0;}videoSourcePlayer1.NewFrame += VideoSourcePlayer1_NewFrame;// 开发者在百度AI平台人脸识别接口只能1秒中调用2次，所以需要做 定时开始检测，每个一秒检测2次ThreadPool.QueueUserWorkItem(new WaitCallback(p => {while (true){IsStart = true;Thread.Sleep(500);}}));}

其次，在NewFrame的回调方法中，根据IsStart判断是否要开始人脸识别，并且另外启动一个线程进行人脸识别操作，判断如果已经有识别过的结构，根据返回的人脸的位置，在当前的一帧图像中绘制方框指示出识别出的人脸位置

private void VideoSourcePlayer1_NewFrame(object sender, ref Bitmap image){try{if (IsStart){IsStart = false;// 在线程池中另起一个线程进行人脸检测,这样不会造成界面视频卡顿现象ThreadPool.QueueUserWorkItem(new WaitCallback(this.Detect), image.Clone());}if (location != null){try{// 绘制方框套住人脸Graphics g = Graphics.FromImage(image);g.DrawLine(new Pen(Color.Black), new System.Drawing.Point(location.left, location.top), new System.Drawing.Point(location.left + location.width, location.top));g.DrawLine(new Pen(Color.Black), new System.Drawing.Point(location.left, location.top), new System.Drawing.Point(location.left, location.top + location.height));g.DrawLine(new Pen(Color.Black), new System.Drawing.Point(location.left, location.top + location.height), new System.Drawing.Point(location.left + location.width, location.top + location.height));g.DrawLine(new Pen(Color.Black), new System.Drawing.Point(location.left + location.width, location.top), new System.Drawing.Point(location.left + location.width, location.top + location.height));g.Dispose();}catch (Exception ex){ClassLoger.Error("VideoSourcePlayer1_NewFrame", ex);}}} catch (Exception ex){ClassLoger.Error("VideoSourcePlayer1_NewFrame1", ex);}}

人脸注册。

在一些类似刷脸签到、刷脸登录的应用场景中，根据人脸获取人物信息，前提就是人脸注册，人脸注册就是获取当前摄像头的一帧图像，调用百度AI的人脸注册接口进行注册

// 用户IDstring uid = "1";// 用户资料，长度限制256Bstring userInfo = textBox6.Text.Trim();// 用户组IDstring groupId = textBox5.Text.Trim();if (comboBox1.Items.Count <= 0){MessageBox.Show("请插入视频设备");return;}try{if (videoSourcePlayer1.IsRunning){BitmapSource bitmapSource = System.Windows.Interop.Imaging.CreateBitmapSourceFromHBitmap(videoSourcePlayer1.GetCurrentVideoFrame().GetHbitmap(),IntPtr.Zero,Int32Rect.Empty,BitmapSizeOptions.FromEmptyOptions());var img = BitmapSource2Byte(bitmapSource);var options = new Dictionary<string, object>{{"action_type", "replace"}};var result = client.UserAdd(uid, userInfo, groupId, img, options);if (result.ToString().Contains("error_code")){MessageBox.Show("注册失败:" + result.ToString());}else{MessageBox.Show("注册成功");}}}catch (Exception ex){MessageBox.Show("摄像头异常：" + ex.Message);}

人脸登录

人脸登录和人脸注册的方式一样，只不过调用的是百度AI的人脸登录接口

// 用户IDstring uid = "1";// 用户资料，长度限制256Bstring userInfo = textBox6.Text.Trim();// 用户组IDstring groupId = textBox5.Text.Trim();if (comboBox1.Items.Count <= 0){MessageBox.Show("请插入视频设备");return;}try{if (videoSourcePlayer1.IsRunning){BitmapSource bitmapSource = System.Windows.Interop.Imaging.CreateBitmapSourceFromHBitmap(videoSourcePlayer1.GetCurrentVideoFrame().GetHbitmap(),IntPtr.Zero,Int32Rect.Empty,BitmapSizeOptions.FromEmptyOptions());var img = BitmapSource2Byte(bitmapSource);// 如果有可选参数//var options = new Dictionary<string, object>{//    {"ext_fields", "faceliveness"},//    {"user_top_num", 3}//};var result = client.Identify(groupId, img);FaceIdentifyInfo info = JsonHelper.DeserializeObject<FaceIdentifyInfo>(result.ToString());if (info!=null && info.result!=null && info.result.Length>0){textBox7.Text = info.result[0].user_info;}}}catch (Exception ex){MessageBox.Show("摄像头异常：" + ex.Message);}

源码地址：https://github.com/liemei/baiduAIFaceIdentify

转载于:https://www.cnblogs.com/liemei/p/8318414.html