H264-YUV通过RTP接收视频流ffmpeg解码SDL实时播放

写在前面的话

　　写一个简单的播放器，通过RTP接收视频流，进行实时播放。最初，使用ffplay或者vlc接收按照SDP协议文件可以播放视频，但是视频中断后重启，不能正确的解包，时常会出现如下的错误信息。

[sdp @ 0x7ffb35034e00] RTP: dropping old packet received too late
    Last message repeated 15 times

　　

使用ffplay播放udp视频。

➜  view-show ffplay -protocol_whitelist "file,http,https,rtp,udp,tcp,tls" test.sdp
➜  view-show cat test.sdp
m=video 6665 RTP/AVP 96
a=rtpmap:96 H264/90000
c=IN IP4 0.0.0.0

　　分析ffmpeg原代码，在rtpdec.c代码文件中，这个错误的原因是ffplay接收RTP视频流时，如果前一个RTP包的流水号大于后一个帧时，会将当前RTP报文丢弃。

static int rtp_parse_one_packet(RTPDemuxContext *s, AVPacket *pkt,                                uint8_t **bufptr, int len)
{
...
    if ((s->seq == 0 && !s->queue) || s->queue_size <= 1) {
        /* First packet, or no reordering */
        return rtp_parse_packet_internal(s, pkt, buf, len);
    } else {
        uint16_t seq = AV_RB16(buf + 2);
        int16_t diff = seq - s->seq;
        if (diff < 0) {
            /* 注意看这里 Packet older than the previously emitted one, drop */

            av_log(s->ic, AV_LOG_WARNING,
                   "RTP: dropping old packet received too late\n");
            return -1;
        } else if (diff <= 1) {
            /* Correct packet */
            rv = rtp_parse_packet_internal(s, pkt, buf, len);
            return rv;
        } else {
            /* Still missing some packet, enqueue this one. */
            rv = enqueue_packet(s, buf, len);
            if (rv < 0)
                return rv;
            *bufptr = NULL;
            /* Return the first enqueued packet if the queue is full,
             * even if we're missing something */
            if (s->queue_len >= s->queue_size) {
                av_log(s->ic, AV_LOG_WARNING, "jitter buffer full\n");
                return rtp_parse_queued_packet(s, pkt);
            }
            return -1;
        }
    }
}

　　

但是，实际的业务场合中，对于一个大的视频文件，会按照MTU（以太网1500）拆分成很多个RTP报文（1400大小），多帧视频拆包的个数远超过RTP协议中的流水号限制 65536。

为了绕开这个问题，计划重新做一个简单的视频播放器，功能类似于ffplay接收UDP报文，然后播放。

总体视频传输流程

使用的库简要说明

库	说明和作用
opencv	摄像头或视频文件读取，图像处理
x264	H264视频编码
ffmpeg	H264视频解码
SDL2	视频播放
RTP	RTP视频流打包

细节分析

Opencv-x264-RTP视频编码流程

通过OpenCV读取视频文件代码

void handleVideo(const char* pFileName) {
    Mat frame;
    cv::VideoCapture capture(pFileName);

    while (true) {
        capture >> frame;
        if (frame.empty()) {
            break;
        }

        STREAM_PUSH_INS->push(&frame);
    }
}

通过x264编码视频帧代码

bool X264Encoder::EncodeOneBuf(cv::Mat *yuvMat, Str *resStr) {
    TimeMeasurer tm;
    memset(yuv_buffer_, 0, m_width * m_height * 3);

    uint8_t* yuv_buffer =(uint8_t*) yuvMat->data;

    memcpy(picture_in_.img.plane[0], yuv_buffer, m_width*m_height);
    yuv_buffer += m_width*m_height;
    memcpy(picture_in_.img.plane[1], yuv_buffer, m_width*m_height / 4);
    yuv_buffer += m_width*m_height / 4;
    memcpy(picture_in_.img.plane[2], yuv_buffer, m_width*m_height / 4);
    picture_in_.i_type = X264_TYPE_IDR;

    int64_t i_pts = 0;
    picture_in_.i_pts = i_pts++;

    x264_nal_t *nals;
    int nnal;

    int h264size = 0;

    x264_picture_t pic_out;
    x264_picture_init(&pic_out);
    x264_encoder_encode(x264_encoder_, &nals, &nnal, &picture_in_, &pic_out);
    x264_nal_t *nal;
    for (nal = nals; nal < nals + nnal; nal++) {
        memcpy((char*)resStr->data + h264size,nal->p_payload,nal->i_payload);
        h264size = h264size + nal->i_payload;
    }

    resStr->size = h264size;

    LOG_INFO("x264.encode.cost: %lu", tm.Elapsed());
    return true;
}

　　

YUV格式分析

YUV编码中使用IYUV，也叫YUV420p或者I420，

如下是YUV420p的数据格式。

YUV420P分Y，U，V三个分量

U分量紧跟在Y分量之后，接着V分量（即：YUV）

因此数据格式为 YYYY YYYY UU VV

RTP协议定义

typedef struct rtp_header {
    /* little-endian */
    /* byte 0 */
    uint8_t csrc_len:       4;  /* bit: 0~3 */
    uint8_t extension:      1;  /* bit: 4 */
    uint8_t padding:        1;  /* bit: 5*/
    uint8_t version:        2;  /* bit: 6~7 */
    /* byte 1 */
    uint8_t payload_type:   7;  /* bit: 0~6 */
    uint8_t marker:         1;  /* bit: 7 */
    /* bytes 2, 3 */
    uint16_t seq_no;
    /* bytes 4-7 */
    uint32_t timestamp;
    /* bytes 8-11 */
    uint32_t ssrc;
} __attribute__ ((packed)) rtp_header_t; /* 12 bytes */

　　

ffmpeg做H264视频解码分析

ffmpeg做H264解码初始化代码

AVCodec *gCodec = NULL;
AVCodecContext *gCodec_ctx = NULL;
AVCodecParserContext *gParser = NULL;
AVFrame *gAVFrame = NULL;

void doAVCodecInit() {
    avcodec_register(&ff_h264_decoder);
    av_register_codec_parser(&ff_h264_parser);

    gCodec = avcodec_find_decoder(AV_CODEC_ID_H264);
    if (!gCodec) {
        fprintf(stderr, "Codec not found\n");
        exit(1);
    }

    gCodec_ctx = avcodec_alloc_context3(gCodec);
    if (!gCodec_ctx) {
        fprintf(stderr, "Could not allocate video codec context\n");
        exit(1);
    }

    if (avcodec_open2(gCodec_ctx, gCodec, NULL) < 0) {
        fprintf(stderr, "Could not open codec\n");
        exit(1);
    }

    gParser = av_parser_init(AV_CODEC_ID_H264);
    if (!gParser) {
        fprintf(stderr, "Could not create H264 parser\n");
        exit(1);
    }

    gAVFrame = av_frame_alloc();
    if (!gAVFrame) {
        fprintf(stderr, "Could not allocate video frame\n");
        exit(1);
    }

    initPacket();
}

　　

ffmpeg循环做H264解码代码

static int doDecodeFrame(AVPacket *pkt, unsigned int frame_index) {
    int got_frame = 0;
    do {
        int len = avcodec_decode_video2(gCodec_ctx, gAVFrame, &got_frame, pkt);
        if (len < 0) {
            fprintf(stderr, "Error while decoding frame %d\n", frame_index);
            return len;
        }
        if (got_frame) {
            //printf("Got frame %d\n", frame_index);
            //fflush(stdout);
            yuv_show(gAVFrame->data, gAVFrame->linesize, gAVFrame->width, gAVFrame->height);
        }
    } while (0);
    return 0;
}

int doPackDecode(struct ImagePacket *packetPtr) {
    uint8_t *data = NULL;
    int size = 0;
    int bytes_used = av_parser_parse2(gParser, gCodec_ctx, &data, &size, packetPtr->buf_, packetPtr->len_, 0, 0,
                                      AV_NOPTS_VALUE);
    if (size == 0) {
        return -1;
    }

    // We have data of one packet, decode it; or decode whatever when ending
    AVPacket packet;
    av_init_packet(&packet);
    packet.data = data;
    packet.size = size;
    int ret = doDecodeFrame(&packet, packetPtr->frame_index_);
    if (ret < 0) {
        return -1;
    }
    return 0;
}

　　

SDL视频渲染流程

SDL初始化代码

#define LOAD_YUV420P 0

#define HAS_BORDER     1

const int bpp = 12;

const int screen_w = 1434, screen_h = 806;
const int pixel_w = 1434, pixel_h = 806;

//const int screen_w=1920,screen_h=1080;
//const int pixel_w=1920,pixel_h=1080;

SDL_Window *gScreen = NULL;
SDL_Renderer *gSdlRenderer = NULL;
SDL_Texture *gSdlTexture = NULL;
SDL_Rect sdlRect;

//Refresh Event
#define REFRESH_EVENT  (SDL_USEREVENT + 1)

int thread_exit = 0;

int refresh_video(void *opaque) {
    while (thread_exit == 0) {
        SDL_Event event;
        event.type = REFRESH_EVENT;
        SDL_PushEvent(&event);
        SDL_Delay(10);
    }
    return 0;
}

int doSDLInit() {
    if (SDL_Init(SDL_INIT_EVERYTHING)) {
        printf("Could not initialize SDL - %s\n", SDL_GetError());
        return -1;
    }

    //SDL 2.0 Support for multiple windows
    gScreen = SDL_CreateWindow("Video-View", SDL_WINDOWPOS_UNDEFINED, SDL_WINDOWPOS_UNDEFINED,
                               screen_w, screen_h, SDL_WINDOW_OPENGL | SDL_WINDOW_RESIZABLE);
    if (!gScreen) {
        printf("SDL: could not create window - exiting:%s\n", SDL_GetError());
        return -1;
    }
    gSdlRenderer = SDL_CreateRenderer(gScreen, -1, 0);
    int pixformat = SDL_PIXELFORMAT_IYUV;

    gSdlTexture = SDL_CreateTexture(gSdlRenderer, pixformat, SDL_TEXTUREACCESS_STREAMING, pixel_w, pixel_h);

    int border = 0;

    sdlRect.x = 0 + border;
    sdlRect.y = 0 + border;
    sdlRect.w = screen_w - border * 2;
    sdlRect.h = screen_h - border * 2;

    SDL_Thread *refresh_thread = SDL_CreateThread(refresh_video, NULL, NULL);
    return 0;
}

　　

SDL循环渲染代码

    SDL_Event event;
    SDL_WaitEvent(&event);
    if (event.type == REFRESH_EVENT) {
        SDL_UpdateTexture(gSdlTexture, NULL, gImagePacket->buf_, pixel_w);
        SDL_RenderClear(gSdlRenderer);
        SDL_RenderCopy(gSdlRenderer, gSdlTexture, NULL, &sdlRect);
        SDL_RenderPresent(gSdlRenderer);
        //Delay 40ms
        SDL_Delay(1);
    } else if (event.type == SDL_QUIT) {
        _exit(0);
    }

　　

参考：https://blog.csdn.net/leixiaohua1020/article/details/40525591/

发送和接收总的流程如下图所示

视频源码请参考我的github项目。

接收端启动方式：

./viewer 6665

发送端启动方式

/bin/video-pusher 127.0.0.1 6665 ./test-video/d001.mp4

测试视频使用1280*720的视频文件。可以找一段1280*720的视频验证。

视频读取编码和发送：

　　https://github.com/gityf/img-video/tree/master/video/opencv-x264-rtp-pusher

视频接收解密和播放：

　　https://github.com/gityf/img-video/tree/master/video/ffmpeg-h264-sdl-view

结束，祝玩的开心!