• Android打开相机进行人脸识别,使用虹软人脸识别引擎


    上一张效果图,渣画质,能看就好


    功能说明:

    人脸识别使用的是虹软的FreeSDK,包含人脸追踪,人脸检测,人脸识别,年龄、性别检测功能,其中本demo只使用了FT和FR(人脸追踪和人脸识别),封装了开启相机和人脸追踪、识别功能在FaceCameraHelper中。

    实现逻辑:

    打开相机,监听预览数据回调进行人脸追踪,且为每个检测到的人脸都分配一个trackID(上下帧位置变化不大的人脸框可认为是同一个人脸,具体实现的逻辑可见代码),同时,为了人脸搜索,为每个trackID都分配一个状态(识别中,识别失败,识别通过)、姓名,识别通过则在人脸框上显示姓名,否则只显示trackID(本demo没配服务端,只做了模拟操作)。流程说明见下图。


    FaceCameraHelper包含的接口:

    public interface FaceTrackListener {
    
    /**
    * 回传相机预览数据和人脸框位置
    *
    * @param nv21 相机预览数据
    * @param ftFaceList 待处理的人脸列表
    * @param trackIdList 人脸追踪ID列表
    */
    void onPreviewData(byte[] nv21, List<AFT_FSDKFace> ftFaceList, List<Integer> trackIdList);
    
    
    /**
    * 当出现异常时执行
    *
    * @param e 异常信息
    */
    void onFail(Exception e);
    
    
    /**
    * 当相机打开时执行
    *
    * @param camera 相机实例
    */
    void onCameraOpened(Camera camera);
    
    /**
    * 根据自己的需要可以删除部分人脸,比如指定区域、留下最大人脸等
    *
    * @param ftFaceList 人脸列表
    * @param trackIdList 人脸追踪ID列表
    */
    void adjustFaceRectList(List<AFT_FSDKFace> ftFaceList, List<Integer> trackIdList);
    
    /**
    * 请求人脸特征后的回调
    *
    * @param frFace 人脸特征数据
    * @param requestId 请求码
    */
    void onFaceFeatureInfoGet(@Nullable AFR_FSDKFace frFace, Integer requestId);
    }
    
    ```
    FT人脸框绘制并回调数据:
    

      

    @Override
    public void onPreviewFrame(byte[] nv21, Camera camera) {
    if (faceTrackListener != null) {
    ftFaceList.clear();
    int ftCode = ftEngine.AFT_FSDK_FaceFeatureDetect(nv21, previewSize.width, previewSize.height, AFT_FSDKEngine.CP_PAF_NV21, ftFaceList).getCode();
    if (ftCode != 0) {
    faceTrackListener.onFail(new Exception("ft failed,code is " + ftCode));
    }
    refreshTrackId(ftFaceList);
    faceTrackListener.adjustFaceRectList(ftFaceList, currentTrackIdList);
    if (surfaceViewRect != null) {
    Canvas canvas = surfaceViewRect.getHolder().lockCanvas();
    if (canvas == null) {
    faceTrackListener.onFail(new Exception("can not get canvas of surfaceViewRect"));
    return;
    }
    canvas.drawColor(0, PorterDuff.Mode.CLEAR);
    if (ftFaceList.size() > 0) {
    for (int i = 0; i < ftFaceList.size(); i++) {
    Rect adjustedRect = TrackUtil.adjustRect(new Rect(ftFaceList.get(i).getRect()), previewSize.width, previewSize.height, surfaceWidth, surfaceHeight, cameraOrientation, mCameraId);
    TrackUtil.drawFaceRect(canvas, adjustedRect, faceRectColor, faceRectThickness, currentTrackIdList.get(i), nameMap.get(currentTrackIdList.get(i)));
    }
    }
    surfaceViewRect.getHolder().unlockCanvasAndPost(canvas);
    }
    
    faceTrackListener.onPreviewData(nv21, ftFaceList, currentTrackIdList);
    }
    }
    

      


    大多数设备相机预览数据图像的朝向在横屏时为0度。其他情况按逆时针依次增加90度,因此人脸框的绘制需要做同步转化。CameraID为0时,也就是后置摄像头情况,相机预览数据的显示为原画面,而CameraID为1时,也就是前置摄像头情况,相机的预览画面显示为镜像画面,适配的代码:

    /**
         * @param rect          FT人脸框
         * @param previewWidth  相机预览的宽度
         * @param previewHeight 相机预览高度
         * @param canvasWidth   画布的宽度
         * @param canvasHeight  画布的高度
         * @param cameraOri     相机预览方向
         * @param mCameraId     相机ID
         * @return
         */
        static Rect adjustRect(Rect rect, int previewWidth, int previewHeight, int canvasWidth, int canvasHeight, int cameraOri, int mCameraId) {
            if (rect == null) {
                return null;
            }
            if (canvasWidth < canvasHeight) {
                int t = previewHeight;
                previewHeight = previewWidth;
                previewWidth = t;
            }
     
            float horizontalRatio;
            float verticalRatio;
            if (cameraOri == 0 || cameraOri == 180) {
                horizontalRatio = (float) canvasWidth / (float) previewWidth;
                verticalRatio = (float) canvasHeight / (float) previewHeight;
            } else {
                horizontalRatio = (float) canvasHeight / (float) previewHeight;
                verticalRatio = (float) canvasWidth / (float) previewWidth;
            }
            rect.left *= horizontalRatio;
            rect.right *= horizontalRatio;
            rect.top *= verticalRatio;
            rect.bottom *= verticalRatio;
     
            Rect newRect = new Rect();
     
            switch (cameraOri) {
                case 0:
                    if (mCameraId == Camera.CameraInfo.CAMERA_FACING_FRONT) {
                        newRect.left = canvasWidth - rect.right;
                        newRect.right = canvasWidth - rect.left;
                    } else {
                        newRect.left = rect.left;
                        newRect.right = rect.right;
                    }
                    newRect.top = rect.top;
                    newRect.bottom = rect.bottom;
                    break;
                case 90:
                    newRect.right = canvasWidth - rect.top;
                    newRect.left = canvasWidth - rect.bottom;
                    if (mCameraId == Camera.CameraInfo.CAMERA_FACING_FRONT) {
                        newRect.top = canvasHeight - rect.right;
                        newRect.bottom = canvasHeight - rect.left;
                    } else {
                        newRect.top = rect.left;
                        newRect.bottom = rect.right;
                    }
                    break;
                case 180:
                    newRect.top = canvasHeight - rect.bottom;
                    newRect.bottom = canvasHeight - rect.top;
                    if (mCameraId == Camera.CameraInfo.CAMERA_FACING_FRONT) {
                        newRect.left = rect.left;
                        newRect.right = rect.right;
                    } else {
                        newRect.left = canvasWidth - rect.right;
                        newRect.right = canvasWidth - rect.left;
                    }
                    break;
                case 270:
                    newRect.left = rect.top;
                    newRect.right = rect.bottom;
                    if (mCameraId == Camera.CameraInfo.CAMERA_FACING_FRONT) {
                        newRect.top = rect.left;
                        newRect.bottom = rect.right;
                    } else {
                        newRect.top = canvasHeight - rect.right;
                        newRect.bottom = canvasHeight - rect.left;
                    }
                    break;
                default:
                    break;
            }
            return newRect;
        }
    

      


    由于FR引擎不支持多线程调用,因此只能串行执行,若需要更高效的实现,可创建多个FREngine实例进行任务分配。

    FR线程队列:

    private LinkedBlockingQueue<FaceRecognizeRunnable> faceRecognizeRunnables = new LinkedBlockingQueue<FaceRecognizeRunnable>(MAX_FRTHREAD_COUNT);
    

      

    FR线程:

    public class FaceRecognizeRunnable implements Runnable {
    private Rect faceRect;
    private int width;
    private int height;
    private int format;
    private int ori;
    private Integer requestId;
    private byte[]nv21Data;
    public FaceRecognizeRunnable(byte[]nv21Data,Rect faceRect, int width, int height, int format, int ori, Integer requestId) {
    if (nv21Data==null) {
    return;
    }
    this.nv21Data = new byte[nv21Data.length];
    System.arraycopy(nv21Data,0,this.nv21Data,0,nv21Data.length);
    this.faceRect = new Rect(faceRect);
    this.width = width;
    this.height = height;
    this.format = format;
    this.ori = ori;
    this.requestId = requestId;
    }
    
    @Override
    public void run() {
    if (faceTrackListener!=null && nv21Data!=null) {
    if (frEngine != null) {
    AFR_FSDKFace frFace = new AFR_FSDKFace();
    int frCode = frEngine.AFR_FSDK_ExtractFRFeature(nv21Data, width, height, format, faceRect, ori, frFace).getCode();
    if (frCode == 0) {
    faceTrackListener.onFaceFeatureInfoGet(frFace, requestId);
    } else {
    faceTrackListener.onFaceFeatureInfoGet(null, requestId);
    faceTrackListener.onFail(new Exception("fr failed errorCode is " + frCode));
    }
    nv21Data = null;
    }else {
    faceTrackListener.onFaceFeatureInfoGet(null, requestId);
    faceTrackListener.onFail(new Exception("fr failed ,frEngine is null" ));
    }
    if (faceRecognizeRunnables.size()>0){
    executor.execute(faceRecognizeRunnables.poll());
    }
    }
    }
    }
    

      


    上下帧是否为相同人脸的判断(trackID刷新):

    /**
    * 刷新trackId
    *
    * @param ftFaceList 传入的人脸列表
    */
    public void refreshTrackId(List<AFT_FSDKFace> ftFaceList) {
    currentTrackIdList.clear();
    //每项预先填充-1
    for (int i = 0; i < ftFaceList.size(); i++) {
    currentTrackIdList.add(-1);
    }
    //前一次无人脸现在有人脸,填充新增TrackId
    if (formerTrackIdList.size() == 0) {
    for (int i = 0; i < ftFaceList.size(); i++) {
    currentTrackIdList.set(i, ++currentTrackId);
    }
    } else {
    //前后都有人脸,对于每一个人脸框
    for (int i = 0; i < ftFaceList.size(); i++) {
    //遍历上一次人脸框
    for (int j = 0; j < formerFaceRectList.size(); j++) {
    //若是同一张人脸
    if (TrackUtil.isSameFace(SIMILARITY_RECT, formerFaceRectList.get(j), ftFaceList.get(i).getRect())) {
    //记录ID
    currentTrackIdList.set(i, formerTrackIdList.get(j));
    break;
    }
    }
    }
    }
    //上一次人脸框不存在此人脸
    for (int i = 0; i < currentTrackIdList.size(); i++) {
    if (currentTrackIdList.get(i) == -1) {
    currentTrackIdList.set(i, ++currentTrackId);
    }
    }
    formerTrackIdList.clear();
    formerFaceRectList.clear();
    for (int i = 0; i < ftFaceList.size(); i++) {
    formerFaceRectList.add(new Rect(ftFaceList.get(i).getRect()));
    formerTrackIdList.add(currentTrackIdList.get(i));
    }
    }
    

      

    项目地址:https://github.com/wangshengyang1996/FaceTrackDemo

    若有不当的地方望指出。

  • 相关阅读:
    如何用正确的方法写出高质量软件的75条体会(转)
    使用javascript动态添加onclick事件,
    签名和重载
    C#文件后缀名详解
    配置SQL Server 2005 Express的身份验证方式,以及如何启用sa登录名。
    CSS选择符及优先级计算
    关于软件版本的解释
    数据结构形象解释
    CSS属性选择符
    [转载]Repeater三层嵌套
  • 原文地址:https://www.cnblogs.com/Zzz-/p/10572138.html
Copyright © 2020-2023  润新知