这个系列的文章将会研究最纯粹的Android直播的实现,而且不是用现在的集成SDK来达到直播的技术实现,而是从一个比较底层的直播实现来探讨这个技术,这样子对于直播技术的实现,现成的一些直播框架等都有一个比较好的理解。
上一篇文章把Camera的处理以及推流给实现了,但还留下了几个bug,这一篇文章就把一些bug处理一下,主要处理两个bug
直播画面颠倒直播卡顿的问题如果没有看过之前的文章的可以戳这里
首先,先把画面颠倒的问题解决先,颠倒的话,我们可以通过多种方式完成,比如说从Camera里面获取到的NV21数据进行一个旋转的操作也可以,但这里,使用FFmpeg里的filter来完成,顺便学习一下filter的使用
FFmpeg的filter初始化起来非常的复杂,但初始化完成后,使用就非常的简单了。想要了解filter的强大功能,可以看看官方文档
那么我们需要使用filter,那就需要写一个初始化函数了
/** * 初始化filter */int init_filters(const char *filters_descr) { /** * 注册所有AVFilter */ avfilter_register_all(); char args[512]; int ret = 0; AVFilter *buffersrc = avfilter_get_by_name("buffer"); AVFilter *buffersink = avfilter_get_by_name("buffersink"); AVFilterInOut *outputs = avfilter_inout_alloc(); AVFilterInOut *inputs = avfilter_inout_alloc(); enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE }; //为FilterGraph分配内存 filter_graph = avfilter_graph_alloc(); if (!outputs || !inputs || !filter_graph) { ret = AVERROR(ENOMEM); goto end; } /** * 要填入正确的参数 */ snPRintf(args, sizeof(args), "video_size=%dx%d:pix_fmt=%d:time_base=%d/%d:pixel_aspect=%d/%d", src_width, src_height, pCodecCtx->pix_fmt, pCodecCtx->time_base.num, pCodecCtx->time_base.den, pCodecCtx->sample_aspect_ratio.num, pCodecCtx->sample_aspect_ratio.den); //创建并向FilterGraph中添加一个Filter ret = avfilter_graph_create_filter(&buffersrc_ctx, buffersrc, "in", args, NULL, filter_graph); if (ret < 0) { LOGE("Cannot create buffer source/n"); goto end; } //创建并向FilterGraph中添加一个Filter ret = avfilter_graph_create_filter(&buffersink_ctx, buffersink, "out", NULL, NULL, filter_graph); if (ret < 0) { LOGE("Cannot create buffer sink/n"); goto end; } ret = av_opt_set_int_list(buffersink_ctx, "pix_fmts", pix_fmts, AV_PIX_FMT_NONE, AV_OPT_SEARCH_CHILDREN); if (ret < 0) { LOGE("Cannot set output pixel format/n"); goto end; } outputs->name = av_strdup("in"); outputs->filter_ctx = buffersrc_ctx; outputs->pad_idx = 0; outputs->next = NULL; inputs->name = av_strdup("out"); inputs->filter_ctx = buffersink_ctx; inputs->pad_idx = 0; inputs->next = NULL; //将一串通过字符串描述的Graph添加到FilterGraph中 if ((ret = avfilter_graph_parse_ptr(filter_graph, filters_descr, &inputs, &outputs, NULL)) < 0) { LOGE("parse ptr error/n"); goto end; } //检查FilterGraph的配置 if ((ret = avfilter_graph_config(filter_graph, NULL)) < 0) { LOGE("parse config error/n"); goto end; } //缓存frame,用来保存filter后的frame new_frame = av_frame_alloc(); //uint8_t *out_buffer = (uint8_t *) av_malloc(av_image_get_buffer_size(pCodecCtx->pix_fmt, pCodecCtx->width, pCodecCtx->height, 1)); //av_image_fill_arrays(new_frame->data, new_frame->linesize, out_buffer, pCodecCtx->pix_fmt, pCodecCtx->width, pCodecCtx->height, 1); end: avfilter_inout_free(&inputs); avfilter_inout_free(&outputs); return ret;}可以看到filter的初始化是挺麻烦的,但初始化完成后,只需要调用两个函数就可以很方便的使用了
//向FilterGraph中加入一个AVFrame ret = av_buffersrc_add_frame(buffersrc_ctx, yuv_frame); if (ret >= 0) { //从FilterGraph中取出一个AVFrame ret = av_buffersink_get_frame(buffersink_ctx, new_frame); if (ret >= 0) { ret = encode(pCodecCtx, &pkt, new_frame, &got_packet); } else { LOGE("Error while getting the filtergraph/n"); } } else { LOGE("Error while feeding the filtergraph/n"); }所以初始化麻烦,使用起来就很方便了。但是因为进行的旋转的操作,所以旋转后的frame的width和height就设置了,所以要对编码器的宽高进行修改,不然就无法编码成功
到这里,基本上就可以通过filter来把直播画面颠倒的问题给解决掉了。
那么就可以解决第二个问题就是直播卡顿的问题了,这个问题主要是因为pts/dts的设置问题
首先,我们要先把streamerHandle这个native方法修改一下,给它再添加一个参数,这个参数是用于设置pts的
/** * 对每一次预览的数据进行编码推流 * @param data NV21格式的数据 * @param timestamp 用于设置pts * @return 0成功,小于0失败 */ private native int streamerHandle(byte[] data, long timestamp);在LiveActivity里面修改完成后,要刻去更新一下c里面对应的方法,不然就报错了
在这里,为了提高性能,我们可以把Camera的setPreviewCallback换成setPreviewCallbackWithBuffer,这样子就可以避免预览的时候,频繁创建byte[]和频繁的GC
那么把LiveActivity写好之后呢,我们就需要去到native层去设置好pts
av_packet_rescale_ts
这个函数的主要作用就是:将packet中的有效定时字段(timestamp/duration)从一个time_base
转换为另一个time_base
FFmpeg的time_base实际上就是指时间的刻度,
比如说当time_base
为{1, 30}的时候,如果pts为20,
那么要变成time_base
为{1, 1000000}刻度时的pts就要进行转换(20 * 1 / 30) / (1 / 1000000)
而且解码器那里有一个time_base
,编码器又有自己的time_base
,所以当进行操作后,需要进行一个time_base
的转换才行
设置完成这个之后,还需要用传递进来的timestamp计算也pts,并设置好
到这里,就基本上可以把直播画面卡顿的问题给解决掉了。
完整的native代码就如下:
//// Created by Administrator on 2017/2/19.//#include <jni.h>#include <stdio.h>#include <android/log.h>#include "libavcodec/avcodec.h"#include "libavformat/avformat.h"#include "libavutil/time.h"#include "libavutil/imgutils.h"#include "libavfilter/avfiltergraph.h"#include "libavfilter/buffersink.h"#include "libavfilter/buffersrc.h"#include "libavutil/opt.h"#define LOG_TAG "FFmpeg"#define LOGE(format, ...) __android_log_print(ANDROID_LOG_ERROR, LOG_TAG, format, ##__VA_ARGS__)#define LOGI(format, ...) __android_log_print(ANDROID_LOG_INFO, LOG_TAG, format, ##__VA_ARGS__)AVFormatContext *ofmt_ctx = NULL;AVStream *out_stream = NULL;AVPacket pkt;AVCodecContext *pCodecCtx = NULL;AVCodec *pCodec = NULL;AVFrame *yuv_frame;int frame_count;int src_width;int src_height;int y_length;int uv_length;int64_t start_time;/** * 定义filter相关的变量 */const char *filter_descr = "transpose=clock"; //顺时针旋转90度的filter描述AVFilterContext *buffersink_ctx;AVFilterContext *buffersrc_ctx;AVFilterGraph *filter_graph;int filterInitResult;AVFrame *new_frame;/** * 回调函数,用来把FFmpeg的log写到sdcard里面 */void live_log(void *ptr, int level, const char* fmt, va_list vl) { FILE *fp = fopen("/sdcard/123/live_log.txt", "a+"); if(fp) { vfprintf(fp, fmt, vl); fflush(fp); fclose(fp); }}/** * 编码函数 * avcodec_encode_video2被deprecated后,自己封装的 */int encode(AVCodecContext *pCodecCtx, AVPacket* pPkt, AVFrame *pFrame, int *got_packet) { int ret; *got_packet = 0; ret = avcodec_send_frame(pCodecCtx, pFrame); if(ret <0 && ret != AVERROR_EOF) { return ret; } ret = avcodec_receive_packet(pCodecCtx, pPkt); if(ret < 0 && ret != AVERROR(EAGAIN)) { return ret; } if(ret >= 0) { *got_packet = 1; } return 0;}/** * 初始化filter */int init_filters(const char *filters_descr) { /** * 注册所有AVFilter */ avfilter_register_all(); char args[512]; int ret = 0; AVFilter *buffersrc = avfilter_get_by_name("buffer"); AVFilter *buffersink = avfilter_get_by_name("buffersink"); AVFilterInOut *outputs = avfilter_inout_alloc(); AVFilterInOut *inputs = avfilter_inout_alloc(); enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE }; //为FilterGraph分配内存 filter_graph = avfilter_graph_alloc(); if (!outputs || !inputs || !filter_graph) { ret = AVERROR(ENOMEM); goto end; } /** * 要填入正确的参数 */ snprintf(args, sizeof(args), "video_size=%dx%d:pix_fmt=%d:time_base=%d/%d:pixel_aspect=%d/%d", src_width, src_height, pCodecCtx->pix_fmt, pCodecCtx->time_base.num, pCodecCtx->time_base.den, pCodecCtx->sample_aspect_ratio.num, pCodecCtx->sample_aspect_ratio.den); //创建并向FilterGraph中添加一个Filter ret = avfilter_graph_create_filter(&buffersrc_ctx, buffersrc, "in", args, NULL, filter_graph); if (ret < 0) { LOGE("Cannot create buffer source/n"); goto end; } ret = avfilter_graph_create_filter(&buffersink_ctx, buffersink, "out", NULL, NULL, filter_graph); if (ret < 0) { LOGE("Cannot create buffer sink/n"); goto end; } ret = av_opt_set_int_list(buffersink_ctx, "pix_fmts", pix_fmts, AV_PIX_FMT_NONE, AV_OPT_SEARCH_CHILDREN); if (ret < 0) { LOGE("Cannot set output pixel format/n"); goto end; } outputs->name = av_strdup("in"); outputs->filter_ctx = buffersrc_ctx; outputs->pad_idx = 0; outputs->next = NULL; inputs->name = av_strdup("out"); inputs->filter_ctx = buffersink_ctx; inputs->pad_idx = 0; inputs->next = NULL; //将一串通过字符串描述的Graph添加到FilterGraph中 if ((ret = avfilter_graph_parse_ptr(filter_graph, filters_descr, &inputs, &outputs, NULL)) < 0) { LOGE("parse ptr error/n"); goto end; } //检查FilterGraph的配置 if ((ret = avfilter_graph_config(filter_graph, NULL)) < 0) { LOGE("parse config error/n"); goto end; } new_frame = av_frame_alloc(); //uint8_t *out_buffer = (uint8_t *) av_malloc(av_image_get_buffer_size(pCodecCtx->pix_fmt, pCodecCtx->width, pCodecCtx->height, 1)); //av_image_fill_arrays(new_frame->data, new_frame->linesize, out_buffer, pCodecCtx->pix_fmt, pCodecCtx->width, pCodecCtx->height, 1); end: avfilter_inout_free(&inputs); avfilter_inout_free(&outputs); return ret;}JNIEXPORT jstring JNICALLjava_com_xiaoxiao_live_MainActivity_helloFromFFmpeg(JNIEnv *env, jobject instance) { // TODO char info[10000] = {0}; sprintf(info, "%s/n", avcodec_configuration()); return (*env)->NewStringUTF(env, info);}JNIEXPORT jint JNICALLJava_com_xiaoxiao_live_LiveActivity_streamerRelease(JNIEnv *env, jobject instance) { // TODO if(pCodecCtx) { avcodec_close(pCodecCtx); pCodecCtx = NULL; } if(ofmt_ctx) { avio_close(ofmt_ctx->pb); } if(ofmt_ctx) { avformat_free_context(ofmt_ctx); ofmt_ctx = NULL; } if(yuv_frame) { av_frame_free(&yuv_frame); yuv_frame = NULL; } if(filter_graph) { avfilter_graph_free(&filter_graph); filter_graph = NULL; } if(new_frame) { av_frame_free(&new_frame); new_frame = NULL; }}JNIEXPORT jint JNICALLJava_com_xiaoxiao_live_LiveActivity_streamerFlush(JNIEnv *env, jobject instance) { // TODO int ret; int got_packet; AVPacket packet; if(!(pCodec->capabilities & CODEC_CAP_DELAY)) { return 0; } while(1) { packet.data = NULL; packet.size = 0; av_init_packet(&packet); ret = encode(pCodecCtx, &packet, NULL, &got_packet); if(ret < 0) { break; } if(!got_packet) { ret = 0; break; } LOGI("Encode 1 frame size:%d/n", packet.size); AVRational time_base = ofmt_ctx->streams[0]->time_base; AVRational r_frame_rate1 = {60, 2}; AVRational time_base_q = {1, AV_TIME_BASE}; int64_t calc_duration = (double)(AV_TIME_BASE) * (1 / av_q2d(r_frame_rate1)); packet.pts = av_rescale_q(frame_count * calc_duration, time_base_q, time_base); packet.dts = packet.pts; packet.duration = av_rescale_q(calc_duration, time_base_q, time_base); packet.pos = -1; frame_count++; ofmt_ctx->duration = packet.duration * frame_count; ret = av_interleaved_write_frame(ofmt_ctx, &packet); if(ret < 0) { break; } } //写文件尾 av_write_trailer(ofmt_ctx); return 0;}JNIEXPORT jint JNICALLJava_com_xiaoxiao_live_LiveActivity_streamerHandle(JNIEnv *env, jobject instance, jbyteArray data_, jlong timestamp) { jbyte *data = (*env)->GetByteArrayElements(env, data_, NULL); // TODO int ret, i, resultCode; int got_packet = 0; resultCode = 0; /** * 这里就是之前说的NV21转为AV_PIX_FMT_YUV420P这种格式的操作了 */ memcpy(yuv_frame->data[0], data, y_length); for (i = 0; i < uv_length; i++) { *(yuv_frame->data[2] + i) = *(data + y_length + i * 2); *(yuv_frame->data[1] + i) = *(data + y_length + i * 2 + 1); } yuv_frame->format = pCodecCtx->pix_fmt; yuv_frame->width = src_width; yuv_frame->height = src_height; //yuv_frame->pts = frame_count; //yuv_frame->pts = (1.0 / 30) * 90 * frame_count; yuv_frame->pts = timestamp * 30 / 1000000; pkt.data = NULL; pkt.size = 0; av_init_packet(&pkt); if (filterInitResult >= 0) { ret = 0; //向FilterGraph中加入一个AVFrame ret = av_buffersrc_add_frame(buffersrc_ctx, yuv_frame); if (ret >= 0) { //从FilterGraph中取出一个AVFrame ret = av_buffersink_get_frame(buffersink_ctx, new_frame); if (ret >= 0) { ret = encode(pCodecCtx, &pkt, new_frame, &got_packet); } else { LOGE("Error while getting the filtergraph/n"); } } else { LOGE("Error while feeding the filtergraph/n"); } } if(filterInitResult < 0 || ret < 0) { LOGE("encode from yuv data"); /** * 因为通过filter后,packet的宽高已经改变了,初始化的编码器已经无法使用了, * 所以要兼容filter无法初始化的话,需要重新初始化一个对应宽高的编码器 */ //进行编码 //ret = encode(pCodecCtx, &pkt, yuv_frame, &got_packet); } if(ret < 0) { resultCode = -1; LOGE("Encode error/n"); goto end; } if(got_packet) { LOGI("Encode frame: %d/tsize:%d/n", frame_count, pkt.size); frame_count++; pkt.stream_index = out_stream->index; //将packet中的有效定时字段(timestamp/duration)从一个time_base转换为另一个time_base av_packet_rescale_ts(&pkt, pCodecCtx->time_base, out_stream->time_base); //写PTS/DTS /*AVRational time_base1 = ofmt_ctx->streams[0]->time_base; AVRational r_frame_rate1 = {60, 2}; AVRational time_base_q = {1, AV_TIME_BASE}; int64_t calc_duration = (double)(AV_TIME_BASE) * (1 / av_q2d(r_frame_rate1)); pkt.pts = av_rescale_q(frame_count * calc_duration, time_base_q, time_base1); pkt.dts = pkt.pts; pkt.duration = av_rescale_q(calc_duration, time_base_q, time_base1); pkt.pos = -1; //处理延迟 int64_t pts_time = av_rescale_q(pkt.dts, time_base1, time_base_q); int64_t now_time = av_gettime() - start_time; if(pts_time > now_time) { av_usleep(pts_time - now_time); }*/ ret = av_interleaved_write_frame(ofmt_ctx, &pkt); if(ret < 0) { LOGE("Error muxing packet"); resultCode = -1; goto end; } av_packet_unref(&pkt); }end: (*env)->ReleaseByteArrayElements(env, data_, data, 0); return resultCode;}JNIEXPORT jint JNICALLJava_com_xiaoxiao_live_LiveActivity_streamerInit(JNIEnv *env, jobject instance, jint width, jint height) { // TODO int ret = 0; const char *address = "rtmp://192.168.1.102/oflaDemo/test"; src_width = width; src_height = height; //yuv数据格式里面的 y的大小(占用的空间) y_length = width * height; //u/v占用的空间大小 uv_length = y_length / 4; //设置回调函数,写log av_log_set_callback(live_log); //激活所有的功能 av_register_all(); //推流就需要初始化网络协议 avformat_network_init(); //初始化AVFormatContext avformat_alloc_output_context2(&ofmt_ctx, NULL, "flv", address); if(!ofmt_ctx) { LOGE("Could not create output context/n"); return -1; } //寻找编码器,这里用的就是x264的那个编码器了 pCodec = avcodec_find_encoder(AV_CODEC_ID_H264); if(!pCodec) { LOGE("Can not find encoder!/n"); return -1; } //初始化编码器的context pCodecCtx = avcodec_alloc_context3(pCodec); pCodecCtx->pix_fmt = AV_PIX_FMT_YUV420P; //指定编码格式 pCodecCtx->width = height; pCodecCtx->height = width; pCodecCtx->time_base.num = 1; pCodecCtx->time_base.den = 30; pCodecCtx->bit_rate = 800000; pCodecCtx->gop_size = 300; if(ofmt_ctx->oformat->flags & AVFMT_GLOBALHEADER) { pCodecCtx->flags |= CODEC_FLAG_GLOBAL_HEADER; } pCodecCtx->qmin = 10; pCodecCtx->qmax = 51; pCodecCtx->max_b_frames = 3; AVDictionary *dicParams = NULL; av_dict_set(&dicParams, "preset", "ultrafast", 0); av_dict_set(&dicParams, "tune", "zerolatency", 0); //打开编码器 if(avcodec_open2(pCodecCtx, pCodec, &dicParams) < 0) { LOGE("Failed to open encoder!/n"); return -1; } //新建输出流 out_stream = avformat_new_stream(ofmt_ctx, pCodec); if(!out_stream) { LOGE("Failed allocation output stream/n"); return -1; } out_stream->time_base.num = 1; out_stream->time_base.den = 30; //复制一份编码器的配置给输出流 avcodec_parameters_from_context(out_stream->codecpar, pCodecCtx); //打开输出流 ret = avio_open(&ofmt_ctx->pb, address, AVIO_FLAG_WRITE); if(ret < 0) { LOGE("Could not open output URL %s", address); return -1; } ret = avformat_write_header(ofmt_ctx, NULL); if(ret < 0) { LOGE("Error occurred when open output URL/n"); return -1; } //初始化一个帧的数据结构,用于编码用 //指定AV_PIX_FMT_YUV420P这种格式的 yuv_frame = av_frame_alloc(); uint8_t *out_buffer = (uint8_t *) av_malloc(av_image_get_buffer_size(pCodecCtx->pix_fmt, src_width, src_height, 1)); av_image_fill_arrays(yuv_frame->data, yuv_frame->linesize, out_buffer, pCodecCtx->pix_fmt, src_width, src_height, 1); start_time = av_gettime(); /** * 初始化filter */ filterInitResult = init_filters(filter_descr); if(filterInitResult < 0) { LOGE("Filter init error"); } return 0;}那么,到这里就可以把上面说的两个问题给解决掉了,但还是有一定的延迟(还在找原因),而且当摄像头切换成前摄像头的时候,会发现画面还是颠倒的,因为前摄像头需要顺时针旋转270度才行的,那么这时候就会发现filter在处理这个旋转的时候有点局限性了。因为filter初始化太麻烦了,所以用filter来解决这个直播画面颠倒的问题有点麻烦。所以就需要使用另外的方法来解决这个问题了。
那么就应该在编码前就把预览的数据给旋转过来,为了以后后续的扩展,比如说加上美颜功能这些,那就需要在预览前就要对数据进行修改再预览,那就SurfaceView就无法满足这个要求那就需要需要TextureView或GLSurfaceView了。这两个都可以在预览前拿到数据,再自己绘制出来的,GLSurfaceView功能更加强大,所以就可以使用它了
所以现在还有的问题就是:
使用GLSurfaceView解决前后摄像头直播画面颠倒的问题添加声音减少延迟后续功能添加这些问题都需要后续解决的,所以下一篇文章就使用GLSurfaceView来代替filter解决直播画面颠倒的问题
新闻热点
疑难解答