测试环境

首先放一些测试环境，不保证其他环境也能够这样使用：

• Qt库：5.12.2，mscv2019_64
• scrcpy：2.3.1
• FFmpeg：ffmpeg-n5.1.4-1-gae14d9c06b-win64-gpl-shared-5.1
• Adb：34.0.5
• Android环境：MuMu模拟器12

如何启动scrcpy-server

首先需要说明的是，我们是与scrcpy-server建立连接，而单纯想显示手机上的画面与控制，作者github发布有scrcpy.exe可以直接运行使用，而这里我们相当于做另一个scrcpy，从而达到一些自定义控制的目的。与scrcpy-server建立连接，github上开发文档也说明了https://github.com/Genymobile/scrcpy/blob/master/doc/develop.md，这里更详细的说明下与scrcpy-server建立连接的具体细节。为了更好的描述细节，下面所有操作使用Qt代码演示。

1. 连接设备

启动scrcpy-server的所有操作都是经过Adb进行的，不了解Adb命令建议先学习一下相关命令，因此，连接设备前先确保手机上打开了“USB调试”开关。连接设备使用命令adb connect，Qt中执行Adb命令使用QProcess类，这里我们封装一个Adb工具类以方便的执行命令：

//头文件
#pragma once#include <qobject.h>
#include <qprocess.h>/*** @brief Adb命令执行封装类*/
class AdbCommandRunner {
public:explicit AdbCommandRunner(const QString& deviceName = QString());~AdbCommandRunner();/*** @brief 执行Adb命令* @param cmds 参数列表* @param waitForFinished 是否等待执行完成*/void runAdb(const QStringList& cmds, bool waitForFinished = true);/*** @brief 获取执行结果的错误* @return*/QString getLastErr();QString lastFeedback; //执行结果返回的字符串private:QProcess process;QString deviceName;
};//cpp
#include "adbcommandrunner.h"#include <qdebug.h>AdbCommandRunner::AdbCommandRunner(const QString &deviceName): deviceName(deviceName)
{}AdbCommandRunner::~AdbCommandRunner()  {if (process.isOpen()) {process.kill();process.waitForFinished();}
}void AdbCommandRunner::runAdb(const QStringList &cmds, bool waitForFinished) {if (deviceName.isEmpty()) {process.start("adb/adb", cmds);} else {process.start("adb/adb", QStringList({"-s", deviceName}) + cmds);}qDebug() << "do adb execute command:" << "adb " + cmds.join(' ');if (waitForFinished) {process.waitForFinished();}lastFeedback = process.readAllStandardOutput();
}QString AdbCommandRunner::getLastErr() {QString failReason = process.readAllStandardError();if (failReason.isEmpty()) {failReason = lastFeedback;}return failReason;
}

需要注意的是，Adb服务是后台运行的，我们可以直接执行adb connect命令连接设备，adb会自动启动服务，然而启动服务是需要个几秒钟，直接QProcess执行会有个等待时间，正确的做法是，先使用adb start-server启动服务，这个过程可以在线程中执行：

QThread::create([] {QProcess process;process.start("adb/adb", {"start-server"});process.waitForFinished();if (process.exitCode() == 0 && process.exitStatus() == QProcess::NormalExit) {qDebug() << "adb server start finished!";} else {qDebug() << "adb server start failed:" << process.readAll();}
))->start();

如果服务启动成功，并且设备存在，连接时几乎没有等待时间

bool connectToDevice() {AdbCommandRunner runner;runner.runAdb({"connect", deviceAddress});if (runner.lastFeedback.contains("cannot connect to")) {qDebug() << "connect device:" << deviceAddress << "failed, error:" << runner.getLastErr();return false;}qInfo() << "connect device:" << deviceAddress << "success!";return true;
}

2. 推送scrcpy-server到手机上

推送文件自然是使用adb push命令，建议是推送到临时目录/data/local/tmp下：

bool pushServiceToDevice() {auto scrcpyFilePath = QDir::currentPath() + "/scrcpy/scrcpy-server";qDebug() << "scrcpy path:" << scrcpyFilePath;AdbCommandRunner runner;runner.runAdb({"-s", deviceAddress, "push", scrcpyFilePath, "/data/local/tmp/scrcpy-server.jar"});if (!runner.lastFeedback.contains("1 file pushed")) {qDebug() << runner.getLastErr();return false;}return true;
}

3. 建立Adb隧道连接

默认情况下，scrcpy-server是作为客户端，通过adb隧道连接到电脑端的本地Tcp服务器，如开发者文档上描述，这个角色也是可以反转的，只需要在启动服务命令里面添加tunnel_forward=true（注意不是启动scrcpy.exe的命令行参数）。默认角色下，使用adb reverse命令开启隧道连接，需要注意的是，隧道名中需要携带一个8位字符串scid作为标识，这里我们可以使用时间戳代替：

scid = QString::asprintf("%08x", QDateTime::currentSecsSinceEpoch());
AdbCommandRunner runner;
runner.runAdb({"-s", deviceAddress, "reverse", "localabstract:scrcpy_" + scid, "tcp:27183")});

记住这个27183端口，下面使用QTcpServer时正是使用这个端口监听服务的连接。

4. 启动服务

scrcpy-server本身是一个可执行的jar包，启动这个jar包，使用adb shell命令：

serverRunner = new AdbCommandRunner;
QStringList scrcpyServiceOpt;
scrcpyServiceOpt << "-s" << deviceAddress << "shell";
scrcpyServiceOpt << "CLASSPATH=/data/local/tmp/scrcpy-server.jar";
scrcpyServiceOpt << "app_process";
scrcpyServiceOpt << "/";
scrcpyServiceOpt << "com.genymobile.scrcpy.Server";
scrcpyServiceOpt << SCRCPY_VERSION;
scrcpyServiceOpt << "scid=" + scid;
scrcpyServiceOpt << "audio=false"; //不传输音频
scrcpyServiceOpt << "max_fps=" + QString::number(maxFrameRate); //最大帧率
scrcpyServiceOpt << "max_size=1920"; //视频帧最大尺寸
serverRunner->runAdb(scrcpyServiceOpt, false);

需要注意的是，这里QProcess对象需要保存，关闭服务时需要杀死对应的adb shell子进程。在上面参数中scid以及之前的参数是必要的，如果版本号和scid对应不上无法启动服务。更多的控制参数可以参考源代码scrcpy\app\src\server.c第212行开始，其中参数的默认值在scrcpy\app\src\options.c中，启动成功后就会立即通过adb与电脑端本地服务建立连接。

5. 关闭服务

关闭服务时，首先需要结束shell进程，然后关闭隧道即可：

if (serverRunner) {delete serverRunner;serverRunner = nullptr;
}AdbCommandRunner runner;
runner.runAdb({"-s", deviceAddress, "reverse", "--remove", "localabstract:scrcpy_" + scid});

关闭服务之后，scrcpy-server会自己在设备中删除，重新启动服务需要从第2步骤推送文件开始。

使用QTcpServer与scrcpy-server建立连接

上面说了，默认情况下电脑端作为tcp服务器，scrcpy-server作为客户端建立连接，因此，使用QTcpServer监听本地adb隧道连接端口即可：

ScrcpyServer::ScrcpyServer(QObject *parent): QObject(parent)
{//tcp服务tcpServer = new QTcpServer(this);connect(tcpServer, &QTcpServer::acceptError, this, [] (QAbstractSocket::SocketError socketError) {qCritical() << "scrcpy server accept error:" << socketError;});connect(tcpServer, &QTcpServer::newConnection, this, &ScrcpyServer::handleNewConnection);
}void ScrcpyServer::handleNewConnection() {auto socket = tcpServer->nextPendingConnection();//第一个socket为视频流if (!videoSocket) {videoSocket = socket;connect(socket, &QTcpSocket::readyRead, this, &ScrcpyServer::receiveVideoBuffer);qInfo() << "video socket pending connect...";} else if (!controlSocket) {controlSocket = socket;connect(socket, &QTcpSocket::readyRead, this, &ScrcpyServer::receiveControlBuffer);qInfo() << "control socket pending connect...";} else {qWarning() << "unexpect socket appending...";}connect(socket, &QTcpSocket::stateChanged, this, [=] (QAbstractSocket::SocketState state) {qDebug() << "socket state changed:" << state;if (state == QAbstractSocket::UnconnectedState) {socket->deleteLater();}});
}bool ScrcpyServer::start() {if (!tcpServer->isListening()) {bool success = tcpServer->listen(QHostAddress::AnyIPv4, 27183);if (!success) {qDebug() << "tcp server listen failed:" << tcpServer->errorString();}}
}

根据开发者文档描述，scrcpy-server连接到QTcpServer后，会有3个tcp连接分别用来传输：视频、音频、控制命令，这里我们在启动时设置了audio=false关闭了音频传输，因此第2个为控制socket。

建立连接并视频推流完整流程

上面讲了启动scrcpy-server和使用QTcpServer建立连接，事实上，建立连接和启动tcp服务是需要按照顺序进行的：

1. 开启视频推流过程

• 开启QTcpServer服务，监听指定端口如27183
• 推送scrcpy-server到手机上
• 使用tcp服务监听的端口，和8位随机字符串作为scid，建立Adb隧道连接
• 使用adb shell命令启动scrcpy-server服务
• QTcpServer等待视频流和控制socket连接

2. 关闭视频推流过程

• 结束adb shell子进程
• 关闭Adb隧道连接
• 关闭Tcp服务

视频流的解码

1. 数据包协议解析

文档中详细描述了视频流的数据组成，最开始视频流会传输64字节表示设备的名称，然后依次传输4字节编码方式、4字节帧图像宽度、4字节帧图像高度，接着开始传输视频帧，其中视频帧由帧头和数据组成，帧头中包含有PTS标志（8字节）和帧数据长度（4字节）两个信息，后面接收帧数据长度的数据即可，然后等待接收下一帧数据。视频默认编码为H.264，可以通过启动服务参数更改编码类型，这里我们使用FFmpeg来解析视频帧。
由于解码是个耗时任务，需要放到线程中运行，这里就需要与QTcpSocket接收到的数据进行线程同步处理，为了让解码线程看起来像是以同步方式读取数据，编写一个工具类来接收QTcpSocket发送来的数据：

//头文件
#pragma once#include <qobject.h>
#include <qmutex.h>
#include <qwaitcondition.h>#include "byteutil.h"class BufferReceiver : public QObject {
public:explicit BufferReceiver(QObject *parent = nullptr);void sendBuffer(const QByteArray& data);void endCache();template<typename T>T receive() {enum {T_Size = sizeof(T)};T value = T();receive((void*)&value, T_Size);ByteUtil::swapBits(value);return value;}void receive(void* data, int len);bool isEndReceive() const {return endBufferCache;}private:QByteArray receiveBuffer;QMutex mutex;QWaitCondition receiveWait;bool endBufferCache;
};//cpp
#include "bufferreceiver.h"BufferReceiver::BufferReceiver(QObject *parent): QObject(parent), endBufferCache(false)
{}void BufferReceiver::sendBuffer(const QByteArray &data) {QMutexLocker locker(&mutex);receiveBuffer.append(data);receiveWait.notify_all();
}void BufferReceiver::endCache() {QMutexLocker locker(&mutex);endBufferCache = true;receiveWait.notify_all();
}void BufferReceiver::receive(void *data, int len) {mutex.lock();if (endBufferCache) {mutex.unlock();return;}while (receiveBuffer.size() < len && !endBufferCache) {receiveWait.wait(&mutex);}if (!endBufferCache) {memcpy(data, receiveBuffer.data(), len);receiveBuffer = receiveBuffer.mid(len);}mutex.unlock();
}

在主线程中收到视频流数据就缓存到BufferReceiver中：

void ScrcpyServer::receiveVideoBuffer() {if (videoDecoder) {videoDecoder->appendBuffer(videoSocket->readAll());}
}

解码器线程按照协议依次接收数据包：

void VideoDecoder::run() {QByteArray remoteDeviceName(64, '\0');bufferReceiver.receive(remoteDeviceName.data(), remoteDeviceName.size());auto name = QString::fromUtf8(remoteDeviceName);if (!name.isEmpty()) {qInfo() << "device name received:" << name;}if (bufferReceiver.isEndReceive()) {return;}if (codecCtx == nullptr) {auto codecId = bufferReceiver.receive<uint32_t>();auto width = bufferReceiver.receive<int>();auto height = bufferReceiver.receive<int>();if (!codecInit(codecId, width, height)) {codecRelease();qCritical() << "video decode init failed!";return;}}qInfo() << "video decode is running...";for (;;) {if (!frameReceive()) {break;}if (!frameMerge()) {av_packet_unref(packet);break;}frameUnpack();av_packet_unref(packet);}//释放资源codecRelease();qInfo() << "video decoder exit...";
}

2. 解码流程

注意上面解码线程的读取数据步骤，在读取到解码器和帧大小时就可以进行解码器初始化了：

//初始化解码器
auto codec = avcodec_find_decoder(AV_CODEC_ID_H264);
if (!codec) {qDebug() << "find codec h264 fail!";return false;
}//初始化解码器上下文
codecCtx = avcodec_alloc_context3(codec);
if (!codecCtx) {qDebug() << "allocate codec context fail!";return false;
}codecCtx->width = width;
codecCtx->height = height;
codecCtx->pix_fmt = AV_PIX_FMT_YUV420P;int ret = avcodec_open2(codecCtx, codec, nullptr);
if (ret < 0) {qDebug() << "open codec fail!";return false;
}packet = av_packet_alloc();
if (!packet) {qDebug() << "alloc packet fail!";return false;
}decodeFrame = av_frame_alloc();
if (!decodeFrame) {qDebug() << "alloc frame fail!";return false;
}

获取到帧数据时，依次读取PTS和帧数据大小，设置到AVPacket中：

bool VideoDecoder::frameReceive() {auto ptsFlags = bufferReceiver.receive<uint64_t>();auto frameLen = bufferReceiver.receive<int32_t>();if (bufferReceiver.isEndReceive()) {return false;}Q_ASSERT(frameLen != 0);if (av_new_packet(packet, frameLen)) {qDebug() << "av new packet failed!";return false;}bufferReceiver.receive(packet->data, frameLen);if (bufferReceiver.isEndReceive()) {return false;}if (ptsFlags & SC_PACKET_FLAG_CONFIG) {packet->pts = AV_NOPTS_VALUE;} else {packet->pts = ptsFlags & SC_PACKET_PTS_MASK;}if (ptsFlags & SC_PACKET_FLAG_KEY_FRAME) {packet->flags |= AV_PKT_FLAG_KEY;}packet->dts = packet->pts;return true;
}

根据PTS判断是否需要进行帧合并：

bool VideoDecoder::frameMerge() {bool isConfig = packet->pts == AV_NOPTS_VALUE;if (isConfig) {free(mergeBuffer);mergeBuffer = (uint8_t*)malloc(packet->size);if (!mergeBuffer) {qDebug() << "merge buffer malloc failed! required size:" << packet->size;return false;}memcpy(mergeBuffer, packet->data, packet->size);mergedSize = packet->size;}else if (mergeBuffer) {if (av_grow_packet(packet, mergedSize)) {qDebug() << "av grow packet failed!";return false;}memmove(packet->data + mergedSize, packet->data, packet->size);memcpy(packet->data, mergeBuffer, mergedSize);free(mergeBuffer);mergeBuffer = nullptr;}return true;
}

视频帧解包分别使用avcodec_send_packet和avcodec_receive_frame，下面代码中演示了如何循环解包，然后转换为QVideoFrame对象（供后面视频渲染使用），注意这里图像格式为YUV420P：

void VideoDecoder::frameUnpack() {if (packet->pts == AV_NOPTS_VALUE) {return;}int ret = avcodec_send_packet(codecCtx, packet);if (ret < 0 && ret != AVERROR(EAGAIN)) {qCritical() << "send packet error:" << ret;} else {//循环解析数据帧for (;;) {ret = avcodec_receive_frame(codecCtx, decodeFrame);if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF) {break;}if (ret) {qCritical() << "could not receive video frame:" << ret;break;}QVideoFrame cachedFrame(codecCtx->width * codecCtx->height * 3 / 2,QSize(codecCtx->width, codecCtx->height),codecCtx->width, QVideoFrame::Format_YUV420P);int imageSize = av_image_get_buffer_size(codecCtx->pix_fmt, codecCtx->width, codecCtx->height, 1);if (cachedFrame.map(QAbstractVideoBuffer::WriteOnly)) {uchar *dstData = cachedFrame.bits();av_image_copy_to_buffer(dstData, imageSize, decodeFrame->data, decodeFrame->linesize,codecCtx->pix_fmt,codecCtx->width, codecCtx->height, 1);cachedFrame.unmap();emit frameDecoded(cachedFrame);}av_frame_unref(decodeFrame);}}
}

3. 视频帧转QImage

有时候我们需要提取视频的一帧图像，例如截图操作，需要直接转RGB图像，这时候有两种方法，一是直接对AVFrame进行转换，也就是上面提到的decodeFrame，使用sws_scale函数，但是需要先初始化一个SwsContext，初始化可以在CodecContext初始化之后进行：

swsContext = sws_getContext(codecCtx->width, codecCtx->height, codecCtx->pix_fmt,codecCtx->width, codecCtx->height, AV_PIX_FMT_RGB24, SWS_BILINEAR,nullptr, nullptr, nullptr);

转换时根据codecCtx信息先构造一个QImage，再调用sws_scale即可：

QImage image = QImage(codecCtx->width, codecCtx->height, QImage::Format_RGB888);
auto imagePtr = image.bits();
//将YUV420p转换为RGB24
const int lineSize[4] = {3*codecCtx->width, 0, 0, 0};
sws_scale(swsContext, (const uint8_t* const*)decodeFrame->data, decodeFrame->linesize,0, codecCtx->height, (uint8_t**)&imagePtr, lineSize);

第二种方法就比较简单了，通过Qt内置的方法转换，正如上面提到，通过av_image_copy_to_buffer函数将AVFrame转成了QVideoFrame最后发送了出来，获取QImage直接调用image函数即可，此时转换出来的格式是ARGB32：

QImage image = cachedFrame.image();

使用OpenGL渲染显示视频流

显示视频最好的办法就是使用OpenGL渲染，这样不会消耗大量的CPU资源，并且原视频帧解码出来的YUV420P也可以在OpenGL中计算。Qt中使用OpenGL自然是继承QOpenGLWidget，Qt官方正好有一个显示视频的控件QVideoWidget，只是没有提供直接设置视频流的方法，仔细阅读Multimedia模块中的QVideoWidget源代码发现，如果使用GLSL，经过QPainterVideoSurface实例，最终进行渲染使用的是QVideoSurfaceGlslPainter，其中支持各种图像帧类型的渲染，其中YUV420P也包含在内，对于YUV420P转RGB使用的是BT709标准。复制源代码中multimediawidgets/qmediaopenglhelper_p.h、multimediawidgets/qpaintervideosurface_p.h、multimediawidgets/qpaintervideosurface.cpp3个文件，自定义一个VideoWidget其中实例化一个QPainterVideoSurface，刷新图片是使用QPainterVideoSurface::present函数即可：

//.h
#pragma once#include <qwidget.h>
#include <qopenglwidget.h>#include "qpaintervideosurface_p.h"class VideoWidget : public QOpenGLWidget {
public:explicit VideoWidget(QWidget *parent = nullptr);~VideoWidget();QPainterVideoSurface *videoSurface() const;QSize sizeHint() const override;public:void setAspectRatioMode(Qt::AspectRatioMode mode);protected:void hideEvent(QHideEvent *event) override;void resizeEvent(QResizeEvent *event) override;void paintEvent(QPaintEvent *event) override;private slots:void formatChanged(const QVideoSurfaceFormat &format);void frameChanged();private:void updateRects();private:QPainterVideoSurface *m_surface;Qt::AspectRatioMode m_aspectRatioMode;QRect m_boundingRect;QRectF m_sourceRect;QSize m_nativeSize;bool m_updatePaintDevice;
};//.cpp
#include "videowidget.h"#include <qevent.h>
#include <qvideosurfaceformat.h>VideoWidget::VideoWidget(QWidget *parent): QOpenGLWidget(parent), m_aspectRatioMode(Qt::KeepAspectRatio), m_updatePaintDevice(true)
{m_surface = new QPainterVideoSurface(this);connect(m_surface, &QPainterVideoSurface::frameChanged, this, &VideoWidget::frameChanged);connect(m_surface, &QPainterVideoSurface::surfaceFormatChanged, this, &VideoWidget::formatChanged);
}QPainterVideoSurface *VideoWidget::videoSurface() const {return m_surface;
}VideoWidget::~VideoWidget() {delete m_surface;
}void VideoWidget::setAspectRatioMode(Qt::AspectRatioMode mode)
{m_aspectRatioMode = mode;updateGeometry();
}QSize VideoWidget::sizeHint() const
{return m_surface->surfaceFormat().sizeHint();
}void VideoWidget::hideEvent(QHideEvent *event)
{m_updatePaintDevice = true;
}void VideoWidget::resizeEvent(QResizeEvent *event)
{updateRects();
}void VideoWidget::paintEvent(QPaintEvent *event)
{QPainter painter(this);if (testAttribute(Qt::WA_OpaquePaintEvent)) {QRegion borderRegion = event->region();borderRegion = borderRegion.subtracted(m_boundingRect);QBrush brush = palette().window();for (const QRect &r : borderRegion)painter.fillRect(r, brush);}if (m_surface->isActive() && m_boundingRect.intersects(event->rect())) {m_surface->paint(&painter, m_boundingRect, m_sourceRect);m_surface->setReady(true);} else {if (m_updatePaintDevice && (painter.paintEngine()->type() == QPaintEngine::OpenGL|| painter.paintEngine()->type() == QPaintEngine::OpenGL2)) {m_updatePaintDevice = false;m_surface->updateGLContext();if (m_surface->supportedShaderTypes() & QPainterVideoSurface::GlslShader) {m_surface->setShaderType(QPainterVideoSurface::GlslShader);} else {m_surface->setShaderType(QPainterVideoSurface::FragmentProgramShader);}}}
}void VideoWidget::formatChanged(const QVideoSurfaceFormat &format)
{m_nativeSize = format.sizeHint();updateRects();updateGeometry();update();
}void VideoWidget::frameChanged()
{update(m_boundingRect);
}void VideoWidget::updateRects()
{QRect rect = this->rect();if (m_nativeSize.isEmpty()) {m_boundingRect = QRect();} else if (m_aspectRatioMode == Qt::IgnoreAspectRatio) {m_boundingRect = rect;m_sourceRect = QRectF(0, 0, 1, 1);} else if (m_aspectRatioMode == Qt::KeepAspectRatio) {QSize size = m_nativeSize;size.scale(rect.size(), Qt::KeepAspectRatio);m_boundingRect = QRect(0, 0, size.width(), size.height());m_boundingRect.moveCenter(rect.center());m_sourceRect = QRectF(0, 0, 1, 1);} else if (m_aspectRatioMode == Qt::KeepAspectRatioByExpanding) {m_boundingRect = rect;QSizeF size = rect.size();size.scale(m_nativeSize, Qt::KeepAspectRatio);m_sourceRect = QRectF(0, 0, size.width() / m_nativeSize.width(), size.height() / m_nativeSize.height());m_sourceRect.moveCenter(QPointF(0.5, 0.5));}
}

开始视频推流之前，初始化Surface，设置使用OpenGL渲染，并指定视频格式为YUV420P：

videoWidget->videoSurface()->setShaderType(QPainterVideoSurface::GlslShader);
videoWidget->videoSurface()->start(QVideoSurfaceFormat(QSize(1920, 1080), QVideoFrame::Format_YUV420P));

从VideoDecoder获取到视频帧时发送到Surface：

connect(decorder, &VideoDecoder::frameDecoded, this, [&](const QVideoFrame& frame) {videoWidget->videoSurface()->present(frame);
});

关闭推流时，同时关闭Surface渲染：

videoWidget->videoSurface()->stop();

控制命令的下发

命令的控制是通过第二个socket发送数据，其命令的编码协议定义和编码在源代码scrcpy\app\src\control_msg.h、scrcpy\app\src\control_msg.c这两个文件中。例如，发送一个点击事件：

namespace ByteUtil {/*** @brief 字节序交换* @tparam T 数值类型* @param data 转换目标数值* @param size 字节序交换大小*/template<typename T>static void swapBits(T& data, size_t size = sizeof(T)) {for (size_t i = 0; i < size / 2; i++) {char* pl = (char*)&data + i;char* pr = (char*)&data + (size - i - 1);if (*pl != *pr) {*pl ^= *pr;*pr ^= *pl;*pl ^= *pr;}}}/*** @brief char*转指定数值类型（大端序）* @tparam T 数值类型* @param data 转换目标数值* @param src 原字节数组* @param srcSize 原字节数组大小*/template<typename T>static void bitConvert(T& data, const void* src, int srcSize = sizeof(T)) {memcpy(&data, src, srcSize);swapBits(data, srcSize);}
}class ControlMsg {
public:static QByteArray injectTouchEvent(android_motionevent_action action, android_motionevent_buttons actionButton,android_motionevent_buttons buttons, uint64_t pointerId,const QSize& screenSize, const QPoint& point, float pressure) {char bytes[32];bytes[0] = SC_CONTROL_MSG_TYPE_INJECT_TOUCH_EVENT;bytes[1] = action;ByteUtil::bitConvert(*(uint64_t*)(bytes + 2), &pointerId);uint32_t x = point.x();ByteUtil::bitConvert(*(uint32_t*)(bytes + 10), &x);uint32_t y = point.y();ByteUtil::bitConvert(*(uint32_t*)(bytes + 14), &y);uint16_t w = screenSize.width();ByteUtil::bitConvert(*(uint16_t*)(bytes + 18), &w);uint16_t h = screenSize.height();ByteUtil::bitConvert(*(uint16_t*)(bytes + 20), &h);uint16_t pressureValue = sc_float_to_u16fp(pressure);ByteUtil::bitConvert(*(uint16_t*)(bytes + 22), &pressureValue);ByteUtil::bitConvert(*(uint32_t*)(bytes + 24), &actionButton);ByteUtil::bitConvert(*(uint32_t*)(bytes + 28), &buttons);return { bytes, 32 };}
};

注册videoWidget事件过滤器，模拟发送鼠标事件：

bool App::eventFilter(QObject *watched, QEvent *event) {if (watched == videoWidget) {if (auto mouseEvent = dynamic_cast<QMouseEvent*>(event)) {auto dstPos = QPoint(qRound(mouseEvent->x() * framePixmapRatio.width()), qRound(mouseEvent->y() * framePixmapRatio.height()));if (mouseEvent->type() == QEvent::MouseButtonPress) {scrcpyServer->sendControl(ControlMsg::injectTouchEvent(AMOTION_EVENT_ACTION_DOWN, AMOTION_EVENT_BUTTON_PRIMARY,AMOTION_EVENT_BUTTON_PRIMARY, 0,frameSrcSize, dstPos, 1.0));} else if (mouseEvent->type() == QEvent::MouseButtonRelease) {scrcpyServer->sendControl(ControlMsg::injectTouchEvent(AMOTION_EVENT_ACTION_UP, AMOTION_EVENT_BUTTON_PRIMARY,AMOTION_EVENT_BUTTON_PRIMARY, 0,frameSrcSize, dstPos, 0.0));} else if (mouseEvent->type() == QEvent::MouseMove) {scrcpyServer->sendControl(ControlMsg::injectTouchEvent(AMOTION_EVENT_ACTION_MOVE, AMOTION_EVENT_BUTTON_PRIMARY,AMOTION_EVENT_BUTTON_PRIMARY, 0,frameSrcSize, dstPos, 1.0));}}}return QObject::eventFilter(watched, event);
}//ScrcpyServer
void ScrcpyServer::sendControl(const QByteArray &controlMsg) {if (controlSocket) {controlSocket->write(controlMsg);}
}

需要注意的是，screenSize参数必须为原视频发送来的图片帧大小，如果界面上的控件进行了缩放，需要按照比例映射到原图片帧位置才能正确的点击。

demo程序的源代码：https://github.com/daonvshu/qt-scrcpyservice