ZLmediakit关于mk_media_input_h264（）输入时间戳的问题

一时失言乱红尘 2022-09-15 11:57 197阅读 0赞

在zlmediakit的mk\_media\_input\_h264（）函数中，我们输入pts和dts如果为0，则代码会自己生成时间戳，但是这个时间戳不平滑，输出视频有卡顿现象，如果输入原始视频流的时间戳，需要将时间戳转化为时间基为\{1，1000\}才可行。  
看代码：

API_EXPORT void API_CALL mk_media_input_h264(mk_media ctx, void *data, int len, uint32_t dts, uint32_t pts) { 
        assert(ctx && data && len > 0);
        MediaHelper::Ptr *obj = (MediaHelper::Ptr *) ctx;
        (*obj)->getChannel()->inputH264((char *) data, len, dts, pts);
    }
    void DevChannel::inputH264(const char *data, int len, uint32_t dts, uint32_t pts) { 
    //作者在这里自己生成时间戳，单位是毫秒，所以我们传入的也是毫秒
        if(dts == 0){ 
            dts = (uint32_t)_aTicker[0].elapsedTime();
        }
        if(pts == 0){ 
            pts = dts;
        }
    
        //由于rtmp/hls/mp4需要缓存时间戳相同的帧，
        //所以使用FrameNoCacheAble类型的帧反而会在转换成FrameCacheAble时多次内存拷贝
        //在此处只拷贝一次，性能开销更低
        auto frame = FrameImp::create<H264Frame>();
        frame->_dts = dts;
        frame->_pts = pts;
        frame->_buffer.assign(data, len);
        frame->_prefix_size = prefixSize(data,len);
        inputFrame(frame);
    }

具体代码在H264RtpEncoder中有体现：

/** * 264 rtp打包类 */
    class H264RtpEncoder : public H264RtpDecoder ,public RtpInfo{ 
    public:
        typedef std::shared_ptr<H264RtpEncoder> Ptr;
    
        /** * @param ssrc ssrc * @param mtu mtu大小 * @param sample_rate 采样率，强制为90000 * @param pt pt类型 * @param interleaved rtsp interleaved */
        H264RtpEncoder(uint32_t ssrc,
                       uint32_t mtu = 1400,
                       uint32_t sample_rate = 90000,//记得这个值
                       uint8_t pt = 96,
                       uint8_t interleaved = TrackVideo * 2);
        ~H264RtpEncoder() { }
    
        /** * 输入264帧 * @param frame 帧数据，必须 */
        void inputFrame(const Frame::Ptr &frame) override;
    
    private:
        void makeH264Rtp(const void *pData, size_t uiLen, bool bMark,  bool gop_pos, uint32_t uiStamp);
    };

H264RtpEncoder::H264RtpEncoder(uint32_t ssrc, uint32_t mtu, uint32_t sample_rate, uint8_t pt, uint8_t interleaved)
            : RtpInfo(ssrc, mtu, sample_rate, pt, interleaved) { 
    }
    
    void H264RtpEncoder::inputFrame(const Frame::Ptr &frame) { 
        auto ptr = frame->data() + frame->prefixSize();
        auto len = frame->size() - frame->prefixSize();
        //frame内部的pts就是我们传入的pts
        auto pts = frame->pts();
        auto nal_type = H264_TYPE(ptr[0]);
        auto packet_size = getMaxSize() - 2;
    
        //末尾5bit为nalu type，固定为28(FU-A)
        auto fu_char_0 = (ptr[0] & (~0x1F)) | 28;
        auto fu_char_1 = nal_type;
        FuFlags *fu_flags = (FuFlags *) (&fu_char_1);
        fu_flags->start_bit = 1;
    
        //超过MTU则按照FU-A模式打包
        if (len > packet_size + 1) { 
            size_t offset = 1;
            while (!fu_flags->end_bit) { 
                if (!fu_flags->start_bit && len <= offset + packet_size) { 
                    //FU-A end
                    packet_size = len - offset;
                    fu_flags->end_bit = 1;
                }
    
                //传入nullptr先不做payload的内存拷贝
                auto rtp = makeRtp(getTrackType(), nullptr, packet_size + 2, fu_flags->end_bit, pts);
                //rtp payload 负载部分
                uint8_t *payload = rtp->getPayload();
                //FU-A 第1个字节
                payload[0] = fu_char_0;
                //FU-A 第2个字节
                payload[1] = fu_char_1;
                //H264 数据
                memcpy(payload + 2, (uint8_t *) ptr + offset, packet_size);
                //输入到rtp环形缓存
                RtpCodec::inputRtp(rtp, fu_flags->start_bit && nal_type == H264Frame::NAL_IDR);
    
                offset += packet_size;
                fu_flags->start_bit = 0;
            }
        } else { 
            //如果帧长度不超过mtu, 则按照Single NAL unit packet per H.264 方式打包
            makeH264Rtp(ptr, len, false, false, pts);
        }
    }
    
    void H264RtpEncoder::makeH264Rtp(const void* data, size_t len, bool mark, bool gop_pos, uint32_t uiStamp) { 
    //makeRtp()这个函数内部会生成pts
        RtpCodec::inputRtp(makeRtp(getTrackType(), data, len, mark, uiStamp), gop_pos);
    }
    
    //具体代码如下：
    RtpPacket::Ptr RtpInfo::makeRtp(TrackType type, const void* data, size_t len, bool mark, uint32_t stamp) { 
        uint16_t payload_len = (uint16_t) (len + RtpPacket::kRtpHeaderSize);
        auto rtp = RtpPacket::create();
        rtp->setCapacity(payload_len + RtpPacket::kRtpTcpHeaderSize);
        rtp->setSize(payload_len + RtpPacket::kRtpTcpHeaderSize);
        rtp->sample_rate = _sample_rate;
        rtp->type = type;
    
        //rtsp over tcp 头
        auto ptr = (uint8_t *) rtp->data();
        ptr[0] = '$';
        ptr[1] = _interleaved;
        ptr[2] = payload_len >> 8;
        ptr[3] = payload_len & 0xFF;
    
        //rtp头
        auto header = rtp->getHeader();
        header->version = RtpPacket::kRtpVersion;
        header->padding = 0;
        header->ext = 0;
        header->csrc = 0;
        header->mark = mark;
        header->pt = _pt;
        header->seq = htons(_seq++);
        //stamp就是当时你传入的时间戳，从这里可以看出你传入的必须是毫秒
        header->stamp = htonl(uint64_t(stamp) * _sample_rate / 1000);
        header->ssrc = htonl(_ssrc);
    
        //有效负载
        if (data) { 
            memcpy(&ptr[RtpPacket::kRtpHeaderSize + RtpPacket::kRtpTcpHeaderSize], data, len);
        }
        return rtp;
    }

另外，rtsp服务器输出的时间戳在打开视频的时候不一定是从0开始的，所以不能用ffmpeg录像，http是从0开始的。

作为一个记录吧