相信大家都对直播不会陌生，直播的技术也越来越成熟了，目前有这样的一个技术，当弹幕飘到主播的脸上的时候，弹幕会自动消失，出了人脸范围内，就继续显示出来。这个原理非常的简单，其实就是人脸识别，将人脸识别范围内的弹幕全都隐藏。说起来容易做起来难，本文将分以下几点讲述如何实现RTMP视频流的人脸识别。

* 方案选择
* ffmpeg的接入
* ffmpeg的数据解析
* OpenGL的数据绘制
* 人脸跟踪以及人脸框的绘制

## 一、方案的选择
笔者一开始想直接使用别人封装好的播放器，输入地址就能播放。接入后发现，确实接入和使用都很简单，也能够显示，但是有一个很致命的问题，就是没有提供获取裸数据的接口，因而没办法进行人脸识别，后面我就转用了ffmpeg。当然如果只是想在设备上播放RTMP流，bilibili的ijkplayer的框架是完全没有问题的，接入和使用都很简单下面是他们的地址。

* https://github.com/Bilibili/ijkplayer

解析方案已经选择完毕，接下来就是绘制和人脸识别，绘制我采用OpenGL。原因是之前有自己封装过一个自定义surfaceView，直接拿来用就可以了。人脸识别引擎我选择虹软的引擎，原因有二，一是使用起来比较简单，虹软的demo写的不错，很多东西可以直接抄过来；二是免费，其他公司的引擎我也用过，都是有试用期限，我不喜欢有期限的东西，而且虹软的引擎效果也不错，当然是我的首选。

## 二、ffmpeg的接入
### 1.目录结构
在src/main目录下新建cpp以及jniLibs目录，并将ffmpeg库放入，如下图所示。

![图片](https://uploader.shimo.im/f/appp0ZTmxQY3MGMW!thumbnail)

### 2.CMakeLists
首先我们在src/main/cpp目录下新建两个文件，CMakeLists.txt，rtmpplayer-lib。CMake用于库间文件的管理与构建，rtmpplayer-lib是放我们解析数据流的jni代码用的。

![图片](https://uploader.shimo.im/f/4W5oTXstrWYpCEcR!thumbnail)

![图片](https://uploader.shimo.im/f/G3Cxc5VIK6svQaHt!thumbnail)

CMakeLists.txt

```
# For more information about using CMake with Android Studio, read the
# documentation: https://d.android.com/studio/projects/add-native-code.html

# Sets the minimum version of CMake required to build the native library.

cmake_minimum_required(VERSION 3.4.1)

# Creates and names a library, sets it as either STATIC
# or SHARED, and provides the relative paths to its source code.
# You can define multiple libraries, and CMake builds them for you.
# Gradle automatically packages shared libraries with your APK.

add_library( # Sets the name of the library.
        rtmpplayer-lib

# Sets the library as a shared library.
        SHARED

# Provides a relative path to your source file(s).
        rtmpplayer-lib.cpp)
include_directories(ffmpeg)

# Searches for a specified prebuilt library and stores the path as a
# variable. Because CMake includes system libraries in the search path by
# default, you only need to specify the name of the public NDK library
# you want to add. CMake verifies that the library exists before
# completing its build.

find_library( # Sets the name of the path variable.
        log-lib

# Specifies the name of the NDK library that
        # you want CMake to locate.
        log)

# Specifies libraries CMake should link to your target library. You
# can link multiple libraries, such as libraries you define in this
# build script, prebuilt third-party libraries, or system libraries.

target_link_libraries( # Specifies the target library.
        rtmpplayer-lib

# Links the target library to the log library
        # included in the NDK.
        ${log-lib}
        android
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libavcodec.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libavdevice.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libavfilter.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libavformat.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libavutil.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libpostproc.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libswresample.so
        ${PROJECT_SOURCE_DIR}/../jniLibs/${CMAKE_ANDROID_ARCH_ABI}/libswscale.so
        )
```
### 3.build.gradle
   我们需要在指定上面我们的CMake文件的位置，以及指定构建的架构。

```
android{
      defaultConfig {
        ...
        ...
        externalNativeBuild {
            cmake {
                abiFilters "armeabi-v7a"
            }
        }
        ndk {
            abiFilters 'armeabi-v7a'    //只生成armv7的so
        }
    }
    externalNativeBuild {
        cmake {
        //path即为上面CMakeLists的地址
            path "src/main/cpp/CMakeLists.txt"
            version "3.10.2"
        }
    }
}
```
### 4.完成构建
  在上述的步骤都完成后，我们就可以构建了，点击build下的refresh linked C++ prject，再点击右侧Gradle/other/externalNativeBuildDebug，等待构建完成后就可以在build/intermediates/cmake下就可以看到自己构建成功的so库了，如果能看到libnative-lib.so那么恭喜你，ffmpeg接入就算完成了。

![图片](https://uploader.shimo.im/f/lDBlrw10GMoNYuQK!thumbnail)

![图片](https://uploader.shimo.im/f/ynLCGFaJmZwXtbWw!thumbnail)

![图片](https://uploader.shimo.im/f/ECtkpA5hXtMGBIS6!thumbnail)

## 三、ffmpeg的数据解析
### 1.JNI数据流解析
上面提到了native-lib.cpp，我们要在这个文件内编写解析RTMP数据流的Jni代码。

```
#include <jni.h>
#include <string>
#include <android/log.h>
#include <fstream>

#define LOGE(FORMAT, ...) __android_log_print(ANDROID_LOG_ERROR, "player", FORMAT, ##__VA_ARGS__);
#define LOGI(FORMAT, ...) __android_log_print(ANDROID_LOG_INFO, "player", FORMAT, ##__VA_ARGS__);

extern "C" {
#include "libavformat/avformat.h"
#include "libavcodec/avcodec.h"
#include "libswscale/swscale.h"
#include "libavutil/imgutils.h"
#include "libavdevice/avdevice.h"
}

static AVPacket *pPacket;
static AVFrame *pAvFrame, *pFrameNv21;
static AVCodecContext *pCodecCtx;
struct SwsContext *pImgConvertCtx;
static AVFormatContext *pFormatCtx;
uint8_t *v_out_buffer;
jobject frameCallback = NULL;
bool stop;
extern "C"
JNIEXPORT jint JNICALL
Java_com_example_rtmpplaydemo_RtmpPlayer_nativePrepare(JNIEnv *env, jobject, jstring url) {
    // 初始化
#if LIBAVCODEC_VERSION_INT < AV_VERSION_INT(55, 28, 1)
#define av_frame_alloc  avcodec_alloc_frame
#endif
    if (frameCallback == NULL) {
        return -1;
    }
    //申请空间
    pAvFrame = av_frame_alloc();
    pFrameNv21 = av_frame_alloc();
    const char* temporary = env->GetStringUTFChars(url,NULL);
    char input_str[500] = {0};
    strcpy(input_str,temporary);
    env->ReleaseStringUTFChars(url,temporary);

//注册库中所有可用的文件格式和编码器
    avcodec_register_all();
    av_register_all();
    avformat_network_init();
    avdevice_register_all();

pFormatCtx = avformat_alloc_context();
    int openInputCode = avformat_open_input(&pFormatCtx, input_str, NULL, NULL);
    LOGI("openInputCode = %d", openInputCode);
    if (openInputCode < 0)
        return -1;
    avformat_find_stream_info(pFormatCtx, NULL);

int videoIndex = -1;
    //遍历各个流，找到第一个视频流,并记录该流的编码信息
    for (unsigned int i = 0; i < pFormatCtx->nb_streams; i++)
    {
        if (pFormatCtx->streams[i]->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
            //这里获取到的videoindex的结果为1.
            videoIndex = i;
            break;
        }
    }
    if (videoIndex == -1) {
        return -1;
    }
    pCodecCtx = pFormatCtx->streams[videoIndex]->codec;
    AVCodec *pCodec = avcodec_find_decoder(pCodecCtx->codec_id);
    avcodec_open2(pCodecCtx, pCodec, NULL);

int width = pCodecCtx->width;
    int height = pCodecCtx->height;
    LOGI("width = %d , height = %d", width, height);
    int numBytes = av_image_get_buffer_size(AV_PIX_FMT_NV21, width, height, 1);
    v_out_buffer = (uint8_t *) av_malloc(numBytes * sizeof(uint8_t));
    av_image_fill_arrays(pFrameNv21->data, pFrameNv21->linesize, v_out_buffer, AV_PIX_FMT_NV21,
                         width,
                         height, 1);
    pImgConvertCtx = sws_getContext(
            pCodecCtx->width,             //原始宽度
            pCodecCtx->height,            //原始高度
            pCodecCtx->pix_fmt,           //原始格式
            pCodecCtx->width,             //目标宽度
            pCodecCtx->height,            //目标高度
            AV_PIX_FMT_NV21,               //目标格式
            SWS_FAST_BILINEAR,                    //选择哪种方式来进行尺寸的改变,关于这个参数,可以参考:http://www.cnblogs.com/mmix2009/p/3585524.html
            NULL,
            NULL,
            NULL);
    pPacket = (AVPacket *) av_malloc(sizeof(AVPacket));
    //onPrepared 回调
    jclass clazz = env->GetObjectClass(frameCallback);
    jmethodID onPreparedId = env->GetMethodID(clazz, "onPrepared", "(II)V");
    env->CallVoidMethod(frameCallback, onPreparedId, width, height);
    env->DeleteLocalRef(clazz);
    return videoIndex;
}

extern "C"
JNIEXPORT void JNICALL
Java_com_example_rtmpplaydemo_RtmpPlayer_nativeStop(JNIEnv *env, jobject) {
    //停止播放
    stop = true;
    if (frameCallback == NULL) {
        return;
    }
    jclass clazz = env->GetObjectClass(frameCallback);
    jmethodID onPlayFinishedId = env->GetMethodID(clazz, "onPlayFinished", "()V");
    //发送onPlayFinished 回调
    env->CallVoidMethod(frameCallback, onPlayFinishedId);
    env->DeleteLocalRef(clazz);
    //释放资源
    sws_freeContext(pImgConvertCtx);
    av_free(pPacket);
    av_free(pFrameNv21);
    avcodec_close(pCodecCtx);
    avformat_close_input(&pFormatCtx);
}

extern "C"
JNIEXPORT void JNICALL
Java_com_example_rtmpplaydemo_RtmpPlayer_nativeSetCallback(JNIEnv *env, jobject,
                                                           jobject callback) {
    //设置回调
    if (frameCallback != NULL) {
        env->DeleteGlobalRef(frameCallback);
        frameCallback = NULL;
    }
    frameCallback = (env)->NewGlobalRef(callback);
}

extern "C"
JNIEXPORT void JNICALL
Java_com_example_rtmpplaydemo_RtmpPlayer_nativeStart(JNIEnv *env, jobject) {
    //开始播放
    stop = false;
    if (frameCallback == NULL) {
        return;
    }
    // 读取数据包
    int count = 0;
    while (!stop) {
        if (av_read_frame(pFormatCtx, pPacket) >= 0) {
            //解码
            int gotPicCount = 0;
            int decode_video2_size = avcodec_decode_video2(pCodecCtx, pAvFrame, &gotPicCount,
                                                           pPacket);
            LOGI("decode_video2_size = %d , gotPicCount = %d", decode_video2_size, gotPicCount);
            LOGI("pAvFrame->linesize  %d  %d %d", pAvFrame->linesize[0], pAvFrame->linesize[1],
                 pCodecCtx->height);
            if (gotPicCount != 0) {
                count++;
                sws_scale(
                        pImgConvertCtx,
                        (const uint8_t *const *) pAvFrame->data,
                        pAvFrame->linesize,
                        0,
                        pCodecCtx->height,
                        pFrameNv21->data,
                        pFrameNv21->linesize);
                //获取数据大小 宽高等数据
                int dataSize = pCodecCtx->height * (pAvFrame->linesize[0] + pAvFrame->linesize[1]);
                LOGI("pAvFrame->linesize  %d  %d %d %d", pAvFrame->linesize[0],
                     pAvFrame->linesize[1], pCodecCtx->height, dataSize);
                jbyteArray data = env->NewByteArray(dataSize);
                env->SetByteArrayRegion(data, 0, dataSize,
                                        reinterpret_cast<const jbyte *>(v_out_buffer));
                // onFrameAvailable 回调
                jclass clazz = env->GetObjectClass(frameCallback);
                jmethodID onFrameAvailableId = env->GetMethodID(clazz, "onFrameAvailable", "([B)V");
                env->CallVoidMethod(frameCallback, onFrameAvailableId, data);
                env->DeleteLocalRef(clazz);
                env->DeleteLocalRef(data);

}
        }
        av_packet_unref(pPacket);
    }
}
```
### 2.Java层数据回调
在上面的jni操作完成后，我们已经获得了解析完成的裸数据，接下来只要将裸数据传到java层，我们也就算是大功告成了，这里我们用回调来实现。

```
//Rtmp回调
public interface PlayCallback {
    //数据准备回调
    void onPrepared(int width, int height);
    //数据回调
    void onFrameAvailable(byte[] data);
    //播放结束回调
    void onPlayFinished();
}

```
接着我们只需要将这个回调传入native，再通过jni将解析好的数据传给java即可。

```
 RtmpPlayer.getInstance().nativeSetCallback(new PlayCallback() {
    @Override
    public void onPrepared(int width, int height) {
        //start 循环调运会阻塞主线程 需要在子线程里运行
        RtmpPlayer.getInstance().nativeStart();
    }

@Override
    public void onFrameAvailable(byte[] data) {
       //获得裸数据，裸数据的格式为NV21
        Log.i(TAG, "onFrameAvailable: " + Arrays.hashCode(data));
        surfaceView.refreshFrameNV21(data);
    }

@Override
    public void onPlayFinished() {
         //播放结束的回调
    }
});
//数据准备
int code = RtmpPlayer.getInstance().prepare("rtmp://58.200.131.2:1935/livetv/hunantv");
if (code == -1) {
    //code为-1则证明rtmp的prepare有问题
    Toast.makeText(MainActivity.this, "prepare Error", Toast.LENGTH_LONG).show();
}
```
  onFrameAvailable得到的data就是我们需要的NV21格式的数据了，下图是我播放湖南卫视得到的数据回调，从hashCode上来看，每次的数据回调都不一样，可以认为数据是实时刷新的。![图片](https://uploader.shimo.im/f/R6Zzkj4mNpIEjTE4!thumbnail)

### 3.Java层与JNI的交互
新建了RtmpPlayer单例类做为Jni与java层交互的通道。

```
public class RtmpPlayer {

private static volatile RtmpPlayer mInstance;
    private static final int PREPARE_ERROR = -1;

private RtmpPlayer() {
    }

//双重锁定防止多线程操作导致的创建多个实例
    public static RtmpPlayer getInstance() {
        if (mInstance == null) {
            synchronized (RtmpPlayer.class) {
                if (mInstance == null) {
                    mInstance = new RtmpPlayer();
                }
            }
        }
        return mInstance;
    }

//数据准备操作
    public int prepare(String url) {
        if(nativePrepare(url) == PREPARE_ERROR){
            Log.i("rtmpPlayer", "PREPARE_ERROR ");
        }
        return nativePrepare(url);
    }

//加载库
    static {
        System.loadLibrary("rtmpplayer-lib");
    }

//数据准备
    private native int nativePrepare(String url);
    //开始播放
    public native void nativeStart();
    //设置回调
    public native void nativeSetCallback(PlayCallback playCallback);
    //停止播放
    public native void nativeStop();
}
```
## 四、小结
至此为止我们已经获得NV21的裸数据，由于时间有限，文章需要写的内容比较多，因此需要分为上下两篇进行讲述。在下篇内我会讲述如何通过OpenGL将我们获得的NV21数据绘制上去，以及如何通过NV21裸数据进行人脸识别，并绘制人脸框。这也是为何我们费尽心机想要得到NV21裸数据的原因。上篇的ffmpeg接入如有问题，可以在下篇最后的附录查看我上传的demo，参考demo可能上手的更容易些。

ITPub博客

Android基于RTMP视频流的人脸识别（上篇）

最新文章