一.概述

本文將在Mac os系統(tǒng)上使用FFmpeg進(jìn)行音視頻的H264，H265編碼。
使用FFmpeg版本為4.2。

二、編碼器初始化

有兩點(diǎn)需要注意的是：
1.設(shè)置pCodecContext->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;的目的是可以通過(guò)pCodecContext->extradata和pCodecContext->extradata_size提取到返回PPS,SPS,VPS數(shù)據(jù)，適用于直播場(chǎng)景，注釋中也寫的很清楚Place global headers in extradata instead of every keyframe.。

如果沒有設(shè)置則會(huì)在AVPacket.data中和其他數(shù)據(jù)一起返回，適用于直接寫入文件。

2.設(shè)置pPacket.flags |= AV_PKT_FLAG_KEY的目的可以在編碼后的AVPacket中識(shí)別出是否為關(guān)鍵幀。

    int ret;
    enum AVCodecID codecID = AV_CODEC_ID_H264;
    if (!kUseH264Encode) {
        codecID = AV_CODEC_ID_HEVC;
    }
    
    pCodec = avcodec_find_encoder(codecID);
    
    pCodecContext = avcodec_alloc_context3(pCodec);
    pCodecContext->codec_type = AVMEDIA_TYPE_VIDEO;
    pCodecContext->pix_fmt = AV_PIX_FMT_YUV420P;
    pCodecContext->width = 1280;
    pCodecContext->height = 720;
    pCodecContext->time_base.num = 1;
    pCodecContext->time_base.den = 25;
    pCodecContext->bit_rate = 1000 * 1000;
    pCodecContext->qmin = 10;
    pCodecContext->qmax = 51;
    pCodecContext->gop_size = 25;
    pCodecContext->max_b_frames = 0;
   // pCodecContext->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
    
    AVDictionary *param = NULL;
    if (kUseH264Encode) {
        av_dict_set(&param, "preset", "slow", 0);
        av_dict_set(&param, "tune", "zerolatency", 0);
    }else{
        av_dict_set(&param, "preset", "ultrafast", 0);
        av_dict_set(&param, "tune", "zero-latency", 0);
    }

    if (avcodec_open2(pCodecContext, pCodec, &param)<0) {
        return;
    }
    
    pFrame = av_frame_alloc();
    pFrame->width = pCodecContext->width;
    pFrame->height = pCodecContext->height;
    pFrame->format = pCodecContext->pix_fmt;
    ret = av_frame_get_buffer(pFrame, 0);
    if (ret < 0) {
        printf("ret == %s\n", av_err2str(ret));
    }
    //初始化avpacket
    av_init_packet(&pPacket);
    pPacket.flags |= AV_PKT_FLAG_KEY;

三、編碼

在Mac OS系統(tǒng)或者iOS系統(tǒng)中，采集到的一般是CMSampleBufferRef對(duì)象，需要先從中拿到CVPixelBufferRef對(duì)象，再?gòu)钠渲刑崛?code>YUV數(shù)據(jù)。而此處的YUV格式，是一種two-plane模式，即Y和UV分為兩個(gè)Plane，但是UV（CbCr）為交錯(cuò)存儲(chǔ)，而不是分為三個(gè)plane，需要最終轉(zhuǎn)換為420P格式，即YYYYUV。

// 鎖定imageBuffer內(nèi)存地址開始進(jìn)行編碼
        if (CVPixelBufferLockBaseAddress(pixelBuffer, 0) == kCVReturnSuccess) {
            //獲取Y分量的地址
            UInt8 *bufferPtr = (UInt8 *)CVPixelBufferGetBaseAddressOfPlane(pixelBuffer,0);
            //獲取UV分量的地址
            UInt8 *bufferPtr1 = (UInt8 *)CVPixelBufferGetBaseAddressOfPlane(pixelBuffer,1);
            //根據(jù)像素獲取圖片的真實(shí)寬度&高度
            size_t width = CVPixelBufferGetWidth(pixelBuffer);
            size_t height = CVPixelBufferGetHeight(pixelBuffer);
            // 獲取Y分量長(zhǎng)度
            size_t bytesrow0 = CVPixelBufferGetBytesPerRowOfPlane(pixelBuffer,0);
            size_t bytesrow1  = CVPixelBufferGetBytesPerRowOfPlane(pixelBuffer,1);
            UInt8 *yuv420_data = (UInt8 *)malloc(width * height * 3 / 2);
            
            //將NV12數(shù)據(jù)轉(zhuǎn)成YUV420P(I420)數(shù)據(jù)
            UInt8 *pY = bufferPtr;
            UInt8 *pUV = bufferPtr1;
            UInt8 *pU = yuv420_data + width * height;
            UInt8 *pV = pU + width * height / 4;
            for(int i =0;i<height;i++)
            {
                memcpy(yuv420_data+i*width,pY+i*bytesrow0,width);
            }
            for(int j = 0;j<height/2;j++)
            {
                for(int i =0;i<width/2;i++)
                {
                    *(pU++) = pUV[i<<1];
                    *(pV++) = pUV[(i<<1) + 1];
                }
                pUV += bytesrow1;
            }
            
            // 3.5.分別讀取YUV的數(shù)據(jù)
            pFrame->data[0] = yuv420_data;
            pFrame->data[1] = pFrame->data[0] + width * height;
            pFrame->data[2] = pFrame->data[1] + (width * height) / 4;
            pFrame->pts = frameCount;
            
            // 5.對(duì)編碼前的原始數(shù)據(jù)(AVFormat)利用編碼器進(jìn)行編碼，將 pFrame 編碼后的數(shù)據(jù)傳入pkt 中
            int ret = avcodec_send_frame(pCodecContext, pFrame);
            if (ret != 0) {
                printf("Failed to encode! \n");
                CVPixelBufferUnlockBaseAddress(pixelBuffer, 0);
                return;
            }
            
            while (1) {
                ret = avcodec_receive_packet(pCodecContext, &pPacket);
                if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF)
                    break;
                else if (ret < 0) {
                    fprintf(stderr, "Error encoding audio frame\n");
                    break;
                }
                frameCount++;
                if (pPacket.flags & AV_PKT_FLAG_KEY) {
                    videoFrame.isKeyFrame =  YES;
                }

                //write file
           NSData *data = [NSData dataWithBytes:pPacket.data length:pPacket.size];
           if ([self.delegate respondsToSelector:@selector(videoEncoder:encodeData:)]) {
              [self.delegate videoEncoder:self encodeData:data];}
                //釋放packet
                av_packet_unref(&pPacket);
            }
            
            // 7.釋放yuv數(shù)據(jù)
            free(yuv420_data);
        }
        CVPixelBufferUnlockBaseAddress(pixelBuffer, 0);
}

四、提取SPS，PPS，VPS數(shù)據(jù)

上文說(shuō)了要單獨(dú)提取SPS，PPS，VPS數(shù)據(jù)，需開始設(shè)置pCodecContext->flags |= AV_CODEC_FLAG_GLOBAL_HEADER。

uint8_t *extra_data = pCodecContext->extradata;
int extra_size = pCodecContext->extradata_size;

1.H264編碼時(shí)，拿到的extra_data如下所示：

00000001 6764001f acb300a0 0b742000 00030020 00000651 e3064d00 00000168 e9732c8b

很明顯SPS，PPS被4個(gè)字節(jié)的start code= 00 00 00 01分割開，NALU header只有一個(gè)字節(jié)：

00 00 00 01 67  ---> (0x67 & 0x1f) = 7 ---> PPS
00 00 00 01 68  ---> (0x68 & 0x1f) = 8 ---> SPS

代碼如下：

int pos = 0;
int pps_pos = 0,pps_length = 0;
int sps_pos = 0,sps_length = 0;
while (pos < (extra_size - 4)) {
      if (extra_data[pos] == 0 &&
         extra_data[pos+1] == 0 &&
         extra_data[pos+2] == 0 &&
         extra_data[pos+3] == 1) {
        if ((extra_data[pos+4] & 0x1f) == 7) {//sps
            sps_pos = pos+4;
        }else if ((extra_data[pos+4] & 0x1f) == 8){//pps
             pps_pos = pos+4;
        }
     }
    pos ++;
}
sps_length = pps_pos - sps_pos - 4;
pps_length = extra_size - pps_pos;

2.H265編碼時(shí)，同樣方法拿到的extra_data提取SPS，PPS，VPS``NALU header有兩個(gè)字節(jié)，提取方法如下：

00 00 00 01 40 01  ---> (0x40 & 0x7E)>>1 = 32 ---> VPS
00 00 00 01 42 01  ---> (0x42 & 0x7E)>>1 = 33 ---> SPS
00 00 00 01 44 01  ---> (0x44 & 0x7E)>>1 = 34 ---> PPS

需要注意的是，此處還可能包含被3個(gè)字節(jié)的start code= 00 00 01分割開的NAL_UNIT_SEI數(shù)據(jù)：

00 00 00 01 4e 01  ---> (0x4e & 0x7E)>>1 = 39 ---> SEI

五、編碼結(jié)束

編碼結(jié)束時(shí)，需要沖洗編碼器，將編碼器中緩存的數(shù)據(jù)沖洗出來(lái)，防止丟幀。方法是發(fā)送avcodec_send_frame(pCodecContext, NULL)，當(dāng)avcodec_receive_packet的返回值為AVERROR_EOF則表示沖洗完成。最后再釋放內(nèi)存。