[OpenCV+FFMepg]使用FFMepg编码cv::Mat并写入视频

OpenCV本身就已经包含很多视频读写的基础能力，但是如果需要更全面的音视频编解码的能力，目前最通用的跨平台方案就是OpenCV+FFMpeg，以下就来介绍如何透过FFMpeg对cv::Mat进行编码，并写入到文件中。

首先，需要引入一些必要头文件，注意FFMpeg的header在C++中引入需要加入extern "C"

#include 
#include 
#include 

#include 
#include 


extern "C" {
#include 
#include 
#include 
#include 
#include 
#include 
}

接著创建一个方法，如下

int writeVideo(const std::string& video_path, std::vector& frames, int width, int height, int fps);

首先需要指定编码的格式，这里以MP4为例，可以根据需求修改

    const AVCodec* videoCodec = avcodec_find_encoder(AV_CODEC_ID_MPEG4);
    AVCodecContext* videoCodecContext = avcodec_alloc_context3(videoCodec);
    if (!videoCodecContext) {
        std::cerr << "Error allocating video codec context" << std::endl;
        exit(EXIT_FAILURE);
    }

接著需要建构编码视频的Context，详细的参数介绍可以参考这里

    videoCodecContext->bit_rate = 200000;
    videoCodecContext->width = width;
    videoCodecContext->height = height;
    //videoCodecContext->time_base = (AVRational){ 1, fps };  //error C4576: a parenthesized type followed by an initializer list is a non-standard explicit type conversion syntax
    //videoCodecContext->framerate = (AVRational){ fps, 1 };
    videoCodecContext->time_base.num = 1;
    videoCodecContext->time_base.den = fps;
    videoCodecContext->framerate.num = fps;
    videoCodecContext->framerate.den = 1;
    videoCodecContext->gop_size = 12;
    videoCodecContext->max_b_frames = 0;
    videoCodecContext->pix_fmt = AV_PIX_FMT_YUV420P;
    if (formatContext->oformat->flags & AVFMT_GLOBALHEADER) {
        videoCodecContext->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
    }

Encoder的参数可以透过这样设置，详细的介绍可以参考这里

// 设置encoder参数
AVDictionary *param = 0;
av_dict_set(¶m, "preset", "medium", 0);
av_dict_set(¶m, "tune", "zerolatency", 0);

//开启Codec并写入设置
error = avcodec_open2(videoCodecContext, videoCodec,  ¶m);

接著就可以开始循环写入cv::Mat

for (const cv::Mat& frame : frames) {
        // convert the cv::Mat to an AVFrame
        AVFrame* avFrame = av_frame_alloc();
        avFrame->format = videoCodecContext->pix_fmt;
        avFrame->width = width;
        avFrame->height = height;
        error = av_frame_get_buffer(avFrame, 0);
        checkError(error, "Error allocating frame buffer");
        struct SwsContext* frameConverter = sws_getContext(width, height, AV_PIX_FMT_BGR24, width, height, videoCodecContext->pix_fmt, SWS_BICUBIC, nullptr, nullptr, nullptr);
        uint8_t* srcData[AV_NUM_DATA_POINTERS] = { frame.data };
        int srcLinesize[AV_NUM_DATA_POINTERS] = { static_cast(frame.step) };
        sws_scale(frameConverter, srcData, srcLinesize, 0, height, avFrame->data, avFrame->linesize);
        sws_freeContext(frameConverter);

...
}

 底下附上完整代码

#include 
#include 
#include 

#include 
#include 


extern "C" {
#include 
#include 
#include 
#include 
#include 
#include 
}

// helper function to check for FFmpeg errors
inline void checkError(int error, const std::string& message) {
    if (error < 0) {
        //std::cerr << message << ": " << av_err2str(error) << std::endl;  //error C4576: a parenthesized type followed by an initializer list is a non-standard explicit type conversion syntax
        std::cerr << message << ": " << std::to_string(error) << std::endl;
        exit(EXIT_FAILURE);
    }
}


int writeVideo(const std::string& video_path, std::vector& frames, int width, int height, int fps) {
    // initialize FFmpeg
    av_log_set_level(AV_LOG_ERROR);
    avformat_network_init();

    // create the output video context
    AVFormatContext* formatContext = nullptr;
    int error = avformat_alloc_output_context2(&formatContext, nullptr, nullptr, video_path.c_str());
    checkError(error, "Error creating output context");

    // create the video stream
    AVStream* videoStream = avformat_new_stream(formatContext, nullptr);
    if (!videoStream) {
        std::cerr << "Error creating video stream" << std::endl;
        exit(EXIT_FAILURE);
    }

    // create the video codec context
    const AVCodec* videoCodec = avcodec_find_encoder(AV_CODEC_ID_MPEG4);
    AVCodecContext* videoCodecContext = avcodec_alloc_context3(videoCodec);
    if (!videoCodecContext) {
        std::cerr << "Error allocating video codec context" << std::endl;
        exit(EXIT_FAILURE);
    }
    videoCodecContext->bit_rate = 200000;
    videoCodecContext->width = width;
    videoCodecContext->height = height;
    //videoCodecContext->time_base = (AVRational){ 1, fps };  //error C4576: a parenthesized type followed by an initializer list is a non-standard explicit type conversion syntax
    //videoCodecContext->framerate = (AVRational){ fps, 1 };
    videoCodecContext->time_base.num = 1;
    videoCodecContext->time_base.den = fps;
    videoCodecContext->framerate.num = fps;
    videoCodecContext->framerate.den = 1;
    videoCodecContext->gop_size = 12;
    videoCodecContext->max_b_frames = 0;
    videoCodecContext->pix_fmt = AV_PIX_FMT_YUV420P;
    if (formatContext->oformat->flags & AVFMT_GLOBALHEADER) {
        videoCodecContext->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
    }

    // 设置encoder参数
    AVDictionary *param = 0;
    av_dict_set(¶m, "preset", "medium", 0);
    av_dict_set(¶m, "tune", "zerolatency", 0);

    error = avcodec_open2(videoCodecContext, videoCodec,  ¶m);
    checkError(error, "Error opening");
    error = avcodec_parameters_from_context(videoStream->codecpar, videoCodecContext);
    checkError(error, "Error setting video codec parameters");

    // open the output file
    error = avio_open(&formatContext->pb, video_path.c_str(), AVIO_FLAG_WRITE);
    checkError(error, "Error opening output file");

    // write the video file header
    error = avformat_write_header(formatContext, nullptr);
    checkError(error, "Error writing video file header");


    AVPacket* packet = av_packet_alloc();
    if (!packet) {
        std::cerr << "Error allocating packet" << std::endl;
        exit(EXIT_FAILURE);
    }
    for (const cv::Mat& frame : frames) {
        // convert the cv::Mat to an AVFrame
        AVFrame* avFrame = av_frame_alloc();
        avFrame->format = videoCodecContext->pix_fmt;
        avFrame->width = width;
        avFrame->height = height;
        error = av_frame_get_buffer(avFrame, 0);
        checkError(error, "Error allocating frame buffer");
        struct SwsContext* frameConverter = sws_getContext(width, height, AV_PIX_FMT_BGR24, width, height, videoCodecContext->pix_fmt, SWS_BICUBIC, nullptr, nullptr, nullptr);
        uint8_t* srcData[AV_NUM_DATA_POINTERS] = { frame.data };
        int srcLinesize[AV_NUM_DATA_POINTERS] = { static_cast(frame.step) };
        sws_scale(frameConverter, srcData, srcLinesize, 0, height, avFrame->data, avFrame->linesize);
        sws_freeContext(frameConverter);

        // encode the AVFrame
        // avFrame->pts = packet->pts;
        avFrame->pts=frame_count*(videoStream->time_base.den)/((videoStream->time_base.num)*fps);
        frame_count += 1;
        error = avcodec_send_frame(videoCodecContext, avFrame);
        checkError(error, "Error sending frame to video codec");
        while (error >= 0) {
            error = avcodec_receive_packet(videoCodecContext, packet);
            if (error == AVERROR(EAGAIN) || error == AVERROR_EOF) {
                break;
            }
            checkError(error, "Error encoding video frame");

            // write the encoded packet to the output file
            packet->stream_index = videoStream->index;
            error = av_interleaved_write_frame(formatContext, packet);
            checkError(error, "Error writing video packet");
            av_packet_unref(packet);
        }
        av_frame_free(&avFrame);
    }


    // flush the rest of the packets
    
    int ret = 0;
    avcodec_send_frame(videoCodecContext, nullptr);
    do
    {
        av_packet_unref(packet);
        ret = avcodec_receive_packet(videoCodecContext, packet);
        if (!ret)
        {
            error = av_interleaved_write_frame(formatContext, packet);
            checkError(error, "Error writing video packet");
        }
    } while (!ret);

    av_write_trailer(formatContext);

    avformat_close_input(&formatContext);
    
   

    // clean up
    av_packet_free(&packet);
    avcodec_free_context(&videoCodecContext);
    avformat_free_context(formatContext);
    avformat_network_deinit();

    return EXIT_SUCCESS;
}

std::vector readVideo(const std::string video_path) {
    // initialize FFmpeg
    av_log_set_level(AV_LOG_ERROR);
    avformat_network_init();

    AVFormatContext* formatContext = nullptr;
    int error = avformat_open_input(&formatContext, video_path.c_str(), nullptr, nullptr);
    checkError(error, "Error opening input file");

    //Read packets of a media file to get stream information.
    
    error = avformat_find_stream_info(formatContext, nullptr);
    checkError(error, "Error avformat find stream info");
    


    // find the video stream
    AVStream* videoStream = nullptr;
    for (unsigned int i = 0; i < formatContext->nb_streams; i++) {
        if (formatContext->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_VIDEO && !videoStream) {
            videoStream = formatContext->streams[i];
        }
    }
    if (!videoStream) {
        std::cerr << "Error: input file does not contain a video stream" << std::endl;
        exit(EXIT_FAILURE);
    }

    // create the video codec context
    const AVCodec* videoCodec = avcodec_find_decoder(videoStream->codecpar->codec_id);
    AVCodecContext* videoCodecContext = avcodec_alloc_context3(videoCodec);
    if (!videoCodecContext) {
        std::cerr << "Error allocating video codec context" << std::endl;
        exit(EXIT_FAILURE);
    }

    std::cout << "::informations::\n";
    std::cout << "  bit_rate:" << videoCodecContext->bit_rate << "\n";
    std::cout << "  width:" << videoCodecContext->width << "\n";
    std::cout << "  height:" << videoCodecContext->height << "\n";
    std::cout << "  gop_size:" << videoCodecContext->gop_size << "\n";
    std::cout << "  max_b_frames:" << videoCodecContext->max_b_frames << "\n";
    std::cout << "  pix_fmt:" << videoCodecContext->pix_fmt << "\n";

    error = avcodec_parameters_to_context(videoCodecContext, videoStream->codecpar);
    checkError(error, "Error setting video codec context parameters");
    error = avcodec_open2(videoCodecContext, videoCodec, nullptr);
    checkError(error, "Error opening video codec");

    // create the frame scaler
    int width = videoCodecContext->width;
    int height = videoCodecContext->height;
    struct SwsContext* frameScaler = sws_getContext(width, height, videoCodecContext->pix_fmt, width, height, AV_PIX_FMT_BGR24, SWS_BICUBIC, nullptr, nullptr, nullptr);

    // read the packets and decode the video frames
    std::vector videoFrames;
    AVPacket packet;
    while (av_read_frame(formatContext, &packet) == 0) {
        if (packet.stream_index == videoStream->index) {
            // decode the video frame
            AVFrame* frame = av_frame_alloc();
            int gotFrame = 0;
            error = avcodec_send_packet(videoCodecContext, &packet);
            checkError(error, "Error sending packet to video codec");
            error = avcodec_receive_frame(videoCodecContext, frame);

            //There is not enough data for decoding the frame, have to free and get more data
            
            if (error == AVERROR(EAGAIN))
            {
                av_frame_unref(frame);
                av_freep(frame);
                continue;
            }

            if (error == AVERROR_EOF)
            {
                std::cerr << "AVERROR_EOF" << std::endl;
                break;
            }
            

            checkError(error, "Error receiving frame from video codec");


            if (error == 0) {
                gotFrame = 1;
            }
            if (gotFrame) {
                // scale the frame to the desired format
                AVFrame* scaledFrame = av_frame_alloc();
                av_image_alloc(scaledFrame->data, scaledFrame->linesize, width, height, AV_PIX_FMT_BGR24, 32);
                sws_scale(frameScaler, frame->data, frame->linesize, 0, height, scaledFrame->data, scaledFrame->linesize);

                // copy the frame data to a cv::Mat object
                cv::Mat mat(height, width, CV_8UC3, scaledFrame->data[0], scaledFrame->linesize[0]);

                //Show mat image for testing
                
                //cv::imshow("mat", mat);
                //cv::waitKey(100);   //Wait 100msec (relativly long time - for testing).
                


                videoFrames.push_back(mat.clone());

                // clean up
                av_freep(&scaledFrame->data[0]);
                av_frame_free(&scaledFrame);
            }
            av_frame_free(&frame);
        }
        av_packet_unref(&packet);
    }


    // clean up
    sws_freeContext(frameScaler);
    avcodec_free_context(&videoCodecContext);
    avformat_close_input(&formatContext);
    return videoFrames;
}

int main() {
    auto videoFrames = readVideo("input.mp4");
    cv::imwrite("test.png", videoFrames[10]);
    writeVideo("outnow.mp4", videoFrames, videoFrames[0].cols, videoFrames[0].rows, 30);