Mirrors/overte-HifiExperiments
3
0
Fork 0
mirror of https://github.com/HifiExperiments/overte.git synced 2025-08-09 12:38:27 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #752 from ey6es/pointy

Browse source 
Depth smoothing, full frame video support (along with ability cycle between no video, just face, full frame).
This commit is contained in:
Philip Rosedale 2013-08-01 21:11:09 -07:00
commit 56e66f8020
7 changed files with 514 additions and 366 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
interface/resources/shaders/face.vert
View file

@ -17,9 +17,6 @@ uniform vec2 texCoordRight;
// the texture coordinate vector from bottom to the top // the texture coordinate vector from bottom to the top
uniform vec2 texCoordUp; uniform vec2 texCoordUp;
// the aspect ratio of the image
uniform float aspectRatio;
// the depth texture // the depth texture
uniform sampler2D depthTexture; uniform sampler2D depthTexture;
@ -31,6 +28,5 @@ void main(void) {
const float MIN_VISIBLE_DEPTH = 1.0 / 255.0; const float MIN_VISIBLE_DEPTH = 1.0 / 255.0;
const float MAX_VISIBLE_DEPTH = 254.0 / 255.0; const float MAX_VISIBLE_DEPTH = 254.0 / 255.0;
gl_FrontColor = vec4(1.0, 1.0, 1.0, step(MIN_VISIBLE_DEPTH, depth) * (1.0 - step(MAX_VISIBLE_DEPTH, depth))); gl_FrontColor = vec4(1.0, 1.0, 1.0, step(MIN_VISIBLE_DEPTH, depth) * (1.0 - step(MAX_VISIBLE_DEPTH, depth)));
gl_Position = gl_ModelViewProjectionMatrix * vec4(0.5 - gl_Vertex.x, gl_Position = gl_ModelViewProjectionMatrix * vec4(0.5 - gl_Vertex.x, gl_Vertex.y - 0.5, depth - 0.5, 1.0);
(gl_Vertex.y - 0.5) / aspectRatio, depth * 2.0 - 2.0, 1.0);
} }

8
interface/src/Application.cpp
View file

@ -965,14 +965,15 @@ void Application::sendAvatarFaceVideoMessage(int frameCount, const QByteArray& d
int headerSize = packetPosition - packet; int headerSize = packetPosition - packet;
// break the data up into submessages of the maximum size // break the data up into submessages of the maximum size (at least one, for zero-length packets)
*offsetPosition = 0; *offsetPosition = 0;
while (*offsetPosition < data.size()) { do {
int payloadSize = min(data.size() - (int)*offsetPosition, MAX_PACKET_SIZE - headerSize); int payloadSize = min(data.size() - (int)*offsetPosition, MAX_PACKET_SIZE - headerSize);
memcpy(packetPosition, data.constData() + *offsetPosition, payloadSize); memcpy(packetPosition, data.constData() + *offsetPosition, payloadSize);
getInstance()->controlledBroadcastToNodes(packet, headerSize + payloadSize, &NODE_TYPE_AVATAR_MIXER, 1); getInstance()->controlledBroadcastToNodes(packet, headerSize + payloadSize, &NODE_TYPE_AVATAR_MIXER, 1);
*offsetPosition += payloadSize; *offsetPosition += payloadSize;
}
} while (*offsetPosition < data.size());
} }
// Every second, check the frame rates and other stuff // Every second, check the frame rates and other stuff
@ -1787,6 +1788,7 @@ void Application::initMenu() {
_testPing->setChecked(true); _testPing->setChecked(true);
(_fullScreenMode = optionsMenu->addAction("Fullscreen", this, SLOT(setFullscreen(bool)), Qt::Key_F))->setCheckable(true); (_fullScreenMode = optionsMenu->addAction("Fullscreen", this, SLOT(setFullscreen(bool)), Qt::Key_F))->setCheckable(true);
optionsMenu->addAction("Webcam", &_webcam, SLOT(setEnabled(bool)))->setCheckable(true); optionsMenu->addAction("Webcam", &_webcam, SLOT(setEnabled(bool)))->setCheckable(true);
optionsMenu->addAction("Cycle Webcam Send Mode", _webcam.getGrabber(), SLOT(cycleVideoSendMode()));
optionsMenu->addAction("Go Home", this, SLOT(goHome())); optionsMenu->addAction("Go Home", this, SLOT(goHome()));
QMenu* renderMenu = menuBar->addMenu("Render"); QMenu* renderMenu = menuBar->addMenu("Render");

495
interface/src/Webcam.cpp
View file

@ -19,6 +19,7 @@
#include "Application.h" #include "Application.h"
#include "Webcam.h" #include "Webcam.h"
#include "avatar/Face.h"
using namespace cv; using namespace cv;
using namespace std; using namespace std;
@ -154,8 +155,8 @@ Webcam::~Webcam() {
const float METERS_PER_MM = 1.0f / 1000.0f; const float METERS_PER_MM = 1.0f / 1000.0f;
void Webcam::setFrame(const Mat& color, int format, const Mat& depth, float meanFaceDepth, void Webcam::setFrame(const Mat& color, int format, const Mat& depth, float midFaceDepth,
const RotatedRect& faceRect, const JointVector& joints) { float aspectRatio, const RotatedRect& faceRect, bool sending, const JointVector& joints) {
IplImage colorImage = color; IplImage colorImage = color;
glPixelStorei(GL_UNPACK_ROW_LENGTH, colorImage.widthStep / 3); glPixelStorei(GL_UNPACK_ROW_LENGTH, colorImage.widthStep / 3);
if (_colorTextureID == 0) { if (_colorTextureID == 0) {
@ -192,8 +193,10 @@ void Webcam::setFrame(const Mat& color, int format, const Mat& depth, float mean
glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
glBindTexture(GL_TEXTURE_2D, 0); glBindTexture(GL_TEXTURE_2D, 0);
// store our face rect and joints, update our frame count for fps computation // store our various data, update our frame count for fps computation
_aspectRatio = aspectRatio;
_faceRect = faceRect; _faceRect = faceRect;
_sending = sending;
_joints = joints; _joints = joints;
_frameCount++; _frameCount++;
@ -242,18 +245,18 @@ void Webcam::setFrame(const Mat& color, int format, const Mat& depth, float mean
if (_initialFaceRect.size.area() == 0) { if (_initialFaceRect.size.area() == 0) {
_initialFaceRect = _faceRect; _initialFaceRect = _faceRect;
_estimatedPosition = glm::vec3(); _estimatedPosition = glm::vec3();
_initialFaceDepth = meanFaceDepth; _initialFaceDepth = midFaceDepth;
} else { } else {
float proportion, z; float proportion, z;
if (meanFaceDepth == UNINITIALIZED_FACE_DEPTH) { if (midFaceDepth == UNINITIALIZED_FACE_DEPTH) {
proportion = sqrtf(_initialFaceRect.size.area() / (float)_faceRect.size.area()); proportion = sqrtf(_initialFaceRect.size.area() / (float)_faceRect.size.area());
const float INITIAL_DISTANCE_TO_CAMERA = 0.333f; const float INITIAL_DISTANCE_TO_CAMERA = 0.333f;
z = INITIAL_DISTANCE_TO_CAMERA * proportion - INITIAL_DISTANCE_TO_CAMERA; z = INITIAL_DISTANCE_TO_CAMERA * proportion - INITIAL_DISTANCE_TO_CAMERA;
} else { } else {
z = (meanFaceDepth - _initialFaceDepth) * METERS_PER_MM; z = (midFaceDepth - _initialFaceDepth) * METERS_PER_MM;
proportion = meanFaceDepth / _initialFaceDepth; proportion = midFaceDepth / _initialFaceDepth;
} }
const float POSITION_SCALE = 0.5f; const float POSITION_SCALE = 0.5f;
_estimatedPosition = glm::vec3( _estimatedPosition = glm::vec3(
@ -270,8 +273,8 @@ void Webcam::setFrame(const Mat& color, int format, const Mat& depth, float mean
QTimer::singleShot(qMax((int)remaining / 1000, 0), _grabber, SLOT(grabFrame())); QTimer::singleShot(qMax((int)remaining / 1000, 0), _grabber, SLOT(grabFrame()));
} }
FrameGrabber::FrameGrabber() : _initialized(false), _capture(0), _searchWindow(0, 0, 0, 0), FrameGrabber::FrameGrabber() : _initialized(false), _videoSendMode(FULL_FRAME_VIDEO), _capture(0), _searchWindow(0, 0, 0, 0),
_smoothedMeanFaceDepth(UNINITIALIZED_FACE_DEPTH), _colorCodec(), _depthCodec(), _frameCount(0) { _smoothedMidFaceDepth(UNINITIALIZED_FACE_DEPTH), _colorCodec(), _depthCodec(), _frameCount(0) {
} }
FrameGrabber::~FrameGrabber() { FrameGrabber::~FrameGrabber() {
@ -364,6 +367,13 @@ static void XN_CALLBACK_TYPE calibrationCompleted(SkeletonCapability& capability
} }
#endif #endif
void FrameGrabber::cycleVideoSendMode() {
_videoSendMode = (VideoSendMode)((_videoSendMode + 1) % VIDEO_SEND_MODE_COUNT);
_searchWindow = cv::Rect(0, 0, 0, 0);
destroyCodecs();
}
void FrameGrabber::reset() { void FrameGrabber::reset() {
_searchWindow = cv::Rect(0, 0, 0, 0); _searchWindow = cv::Rect(0, 0, 0, 0);
@ -379,14 +389,7 @@ void FrameGrabber::shutdown() {
cvReleaseCapture(&_capture); cvReleaseCapture(&_capture);
_capture = 0; _capture = 0;
} }
if (_colorCodec.name != 0) { destroyCodecs();
vpx_codec_destroy(&_colorCodec);
_colorCodec.name = 0;
}
if (_depthCodec.name != 0) {
vpx_codec_destroy(&_depthCodec);
_depthCodec.name = 0;
}
_initialized = false; _initialized = false;
thread()->quit(); thread()->quit();
@ -462,218 +465,208 @@ void FrameGrabber::grabFrame() {
color = image; color = image;
} }
// if we don't have a search window (yet), try using the face cascade int encodedWidth;
int channels = 0; int encodedHeight;
float ranges[] = { 0, 180 }; int depthBitrateMultiplier = 1;
const float* range = ranges; Mat faceTransform;
if (_searchWindow.area() == 0) { float aspectRatio;
vector<Rect> faces; if (_videoSendMode == FULL_FRAME_VIDEO) {
_faceCascade.detectMultiScale(color, faces, 1.1, 6); // no need to find the face if we're sending full frame video
if (!faces.empty()) { _smoothedFaceRect = RotatedRect(Point2f(color.cols / 2.0f, color.rows / 2.0f), Size2f(color.cols, color.rows), 0.0f);
_searchWindow = faces.front(); encodedWidth = color.cols;
updateHSVFrame(color, format); encodedHeight = color.rows;
aspectRatio = FULL_FRAME_ASPECT;
Mat faceHsv(_hsvFrame, _searchWindow);
Mat faceMask(_mask, _searchWindow);
int sizes = 30;
calcHist(&faceHsv, 1, &channels, faceMask, _histogram, 1, &sizes, &range);
double min, max;
minMaxLoc(_histogram, &min, &max);
_histogram.convertTo(_histogram, -1, (max == 0.0) ? 0.0 : 255.0 / max);
}
}
RotatedRect faceRect;
if (_searchWindow.area() > 0) {
updateHSVFrame(color, format);
calcBackProject(&_hsvFrame, 1, &channels, _histogram, _backProject, &range);
bitwise_and(_backProject, _mask, _backProject);
faceRect = CamShift(_backProject, _searchWindow, TermCriteria(CV_TERMCRIT_EPS | CV_TERMCRIT_ITER, 10, 1));
Rect faceBounds = faceRect.boundingRect();
Rect imageBounds(0, 0, color.cols, color.rows);
_searchWindow = Rect(clip(faceBounds.tl(), imageBounds), clip(faceBounds.br(), imageBounds));
}
const int ENCODED_FACE_WIDTH = 128;
const int ENCODED_FACE_HEIGHT = 128;
if (_colorCodec.name == 0) {
// initialize encoder context(s)
vpx_codec_enc_cfg_t codecConfig;
vpx_codec_enc_config_default(vpx_codec_vp8_cx(), &codecConfig, 0);
codecConfig.rc_target_bitrate = ENCODED_FACE_WIDTH * ENCODED_FACE_HEIGHT *
codecConfig.rc_target_bitrate / codecConfig.g_w / codecConfig.g_h;
codecConfig.g_w = ENCODED_FACE_WIDTH;
codecConfig.g_h = ENCODED_FACE_HEIGHT;
vpx_codec_enc_init(&_colorCodec, vpx_codec_vp8_cx(), &codecConfig, 0);
if (!depth.empty()) {
int DEPTH_BITRATE_MULTIPLIER = 2;
codecConfig.rc_target_bitrate *= 2;
vpx_codec_enc_init(&_depthCodec, vpx_codec_vp8_cx(), &codecConfig, 0);
}
}
// correct for 180 degree rotations
if (faceRect.angle < -90.0f) {
faceRect.angle += 180.0f;
} else if (faceRect.angle > 90.0f) {
faceRect.angle -= 180.0f;
}
// compute the smoothed face rect
if (_smoothedFaceRect.size.area() == 0) {
_smoothedFaceRect = faceRect;
} else { } else {
const float FACE_RECT_SMOOTHING = 0.9f; // if we don't have a search window (yet), try using the face cascade
_smoothedFaceRect.center.x = glm::mix(faceRect.center.x, _smoothedFaceRect.center.x, FACE_RECT_SMOOTHING); int channels = 0;
_smoothedFaceRect.center.y = glm::mix(faceRect.center.y, _smoothedFaceRect.center.y, FACE_RECT_SMOOTHING); float ranges[] = { 0, 180 };
_smoothedFaceRect.size.width = glm::mix(faceRect.size.width, _smoothedFaceRect.size.width, FACE_RECT_SMOOTHING); const float* range = ranges;
_smoothedFaceRect.size.height = glm::mix(faceRect.size.height, _smoothedFaceRect.size.height, FACE_RECT_SMOOTHING); if (_searchWindow.area() == 0) {
_smoothedFaceRect.angle = glm::mix(faceRect.angle, _smoothedFaceRect.angle, FACE_RECT_SMOOTHING); vector<Rect> faces;
} _faceCascade.detectMultiScale(color, faces, 1.1, 6);
if (!faces.empty()) {
// resize/rotate face into encoding rectangle _searchWindow = faces.front();
_faceColor.create(ENCODED_FACE_WIDTH, ENCODED_FACE_HEIGHT, CV_8UC3); updateHSVFrame(color, format);
Point2f sourcePoints[4];
_smoothedFaceRect.points(sourcePoints);
Point2f destPoints[] = { Point2f(0, ENCODED_FACE_HEIGHT), Point2f(0, 0), Point2f(ENCODED_FACE_WIDTH, 0) };
Mat transform = getAffineTransform(sourcePoints, destPoints);
warpAffine(color, _faceColor, transform, _faceColor.size());
// convert from RGB to YV12
const int ENCODED_BITS_PER_Y = 8;
const int ENCODED_BITS_PER_VU = 2;
const int ENCODED_BITS_PER_PIXEL = ENCODED_BITS_PER_Y + 2 * ENCODED_BITS_PER_VU;
const int BITS_PER_BYTE = 8;
_encodedFace.resize(ENCODED_FACE_WIDTH * ENCODED_FACE_HEIGHT * ENCODED_BITS_PER_PIXEL / BITS_PER_BYTE);
vpx_image_t vpxImage;
vpx_img_wrap(&vpxImage, VPX_IMG_FMT_YV12, ENCODED_FACE_WIDTH, ENCODED_FACE_HEIGHT, 1, (unsigned char*)_encodedFace.data());
uchar* yline = vpxImage.planes[0];
uchar* vline = vpxImage.planes[1];
uchar* uline = vpxImage.planes[2];
const int Y_RED_WEIGHT = (int)(0.299 * 256);
const int Y_GREEN_WEIGHT = (int)(0.587 * 256);
const int Y_BLUE_WEIGHT = (int)(0.114 * 256);
const int V_RED_WEIGHT = (int)(0.713 * 256);
const int U_BLUE_WEIGHT = (int)(0.564 * 256);
int redIndex = 0;
int greenIndex = 1;
int blueIndex = 2;
if (format == GL_BGR) {
redIndex = 2;
blueIndex = 0;
}
for (int i = 0; i < ENCODED_FACE_HEIGHT; i += 2) {
uchar* ydest = yline;
uchar* vdest = vline;
uchar* udest = uline;
for (int j = 0; j < ENCODED_FACE_WIDTH; j += 2) {
uchar* tl = _faceColor.ptr(i, j);
uchar* tr = _faceColor.ptr(i, j + 1);
uchar* bl = _faceColor.ptr(i + 1, j);
uchar* br = _faceColor.ptr(i + 1, j + 1);
ydest[0] = (tl[redIndex] * Y_RED_WEIGHT + tl[1] * Y_GREEN_WEIGHT + tl[blueIndex] * Y_BLUE_WEIGHT) >> 8; Mat faceHsv(_hsvFrame, _searchWindow);
ydest[1] = (tr[redIndex] * Y_RED_WEIGHT + tr[1] * Y_GREEN_WEIGHT + tr[blueIndex] * Y_BLUE_WEIGHT) >> 8; Mat faceMask(_mask, _searchWindow);
ydest[vpxImage.stride[0]] = (bl[redIndex] * Y_RED_WEIGHT + bl[greenIndex] * int sizes = 30;
Y_GREEN_WEIGHT + bl[blueIndex] * Y_BLUE_WEIGHT) >> 8; calcHist(&faceHsv, 1, &channels, faceMask, _histogram, 1, &sizes, &range);
ydest[vpxImage.stride[0] + 1] = (br[redIndex] * Y_RED_WEIGHT + br[greenIndex] * double min, max;
Y_GREEN_WEIGHT + br[blueIndex] * Y_BLUE_WEIGHT) >> 8; minMaxLoc(_histogram, &min, &max);
ydest += 2; _histogram.convertTo(_histogram, -1, (max == 0.0) ? 0.0 : 255.0 / max);
int totalRed = tl[redIndex] + tr[redIndex] + bl[redIndex] + br[redIndex];
int totalGreen = tl[greenIndex] + tr[greenIndex] + bl[greenIndex] + br[greenIndex];
int totalBlue = tl[blueIndex] + tr[blueIndex] + bl[blueIndex] + br[blueIndex];
int totalY = (totalRed * Y_RED_WEIGHT + totalGreen * Y_GREEN_WEIGHT + totalBlue * Y_BLUE_WEIGHT) >> 8;
*vdest++ = (((totalRed - totalY) * V_RED_WEIGHT) >> 10) + 128;
*udest++ = (((totalBlue - totalY) * U_BLUE_WEIGHT) >> 10) + 128;
}
yline += vpxImage.stride[0] * 2;
vline += vpxImage.stride[1];
uline += vpxImage.stride[2];
}
// encode the frame
vpx_codec_encode(&_colorCodec, &vpxImage, ++_frameCount, 1, 0, VPX_DL_REALTIME);
// start the payload off with the aspect ratio
QByteArray payload(sizeof(float), 0);
*(float*)payload.data() = _smoothedFaceRect.size.width / _smoothedFaceRect.size.height;
// extract the encoded frame
vpx_codec_iter_t iterator = 0;
const vpx_codec_cx_pkt_t* packet;
while ((packet = vpx_codec_get_cx_data(&_colorCodec, &iterator)) != 0) {
if (packet->kind == VPX_CODEC_CX_FRAME_PKT) {
// prepend the length, which will indicate whether there's a depth frame too
payload.append((const char*)&packet->data.frame.sz, sizeof(packet->data.frame.sz));
payload.append((const char*)packet->data.frame.buf, packet->data.frame.sz);
}
}
if (!depth.empty()) {
// warp the face depth without interpolation (because it will contain invalid zero values)
_faceDepth.create(ENCODED_FACE_WIDTH, ENCODED_FACE_HEIGHT, CV_16UC1);
warpAffine(depth, _faceDepth, transform, _faceDepth.size(), INTER_NEAREST);
// find the mean of the valid values
qint64 depthTotal = 0;
qint64 depthSamples = 0;
ushort* src = _faceDepth.ptr<ushort>();
const ushort ELEVEN_BIT_MINIMUM = 0;
const ushort ELEVEN_BIT_MAXIMUM = 2047;
for (int i = 0; i < ENCODED_FACE_HEIGHT; i++) {
for (int j = 0; j < ENCODED_FACE_WIDTH; j++) {
ushort depth = *src++;
if (depth != ELEVEN_BIT_MINIMUM && depth != ELEVEN_BIT_MAXIMUM) {
depthTotal += depth;
depthSamples++;
}
} }
} }
float mean = (depthSamples == 0) ? UNINITIALIZED_FACE_DEPTH : depthTotal / (float)depthSamples; RotatedRect faceRect;
if (_searchWindow.area() > 0) {
updateHSVFrame(color, format);
calcBackProject(&_hsvFrame, 1, &channels, _histogram, _backProject, &range);
bitwise_and(_backProject, _mask, _backProject);
faceRect = CamShift(_backProject, _searchWindow, TermCriteria(CV_TERMCRIT_EPS | CV_TERMCRIT_ITER, 10, 1));
Rect faceBounds = faceRect.boundingRect();
Rect imageBounds(0, 0, color.cols, color.rows);
_searchWindow = Rect(clip(faceBounds.tl(), imageBounds), clip(faceBounds.br(), imageBounds));
}
const int ENCODED_FACE_WIDTH = 128;
const int ENCODED_FACE_HEIGHT = 128;
encodedWidth = ENCODED_FACE_WIDTH;
encodedHeight = ENCODED_FACE_HEIGHT;
depthBitrateMultiplier = 2;
// smooth the mean over time // correct for 180 degree rotations
const float DEPTH_OFFSET_SMOOTHING = 0.95f; if (faceRect.angle < -90.0f) {
_smoothedMeanFaceDepth = (_smoothedMeanFaceDepth == UNINITIALIZED_FACE_DEPTH) ? mean : faceRect.angle += 180.0f;
glm::mix(mean, _smoothedMeanFaceDepth, DEPTH_OFFSET_SMOOTHING);
} else if (faceRect.angle > 90.0f) {
faceRect.angle -= 180.0f;
}
// compute the smoothed face rect
if (_smoothedFaceRect.size.area() == 0) {
_smoothedFaceRect = faceRect;
} else {
const float FACE_RECT_SMOOTHING = 0.9f;
_smoothedFaceRect.center.x = glm::mix(faceRect.center.x, _smoothedFaceRect.center.x, FACE_RECT_SMOOTHING);
_smoothedFaceRect.center.y = glm::mix(faceRect.center.y, _smoothedFaceRect.center.y, FACE_RECT_SMOOTHING);
_smoothedFaceRect.size.width = glm::mix(faceRect.size.width, _smoothedFaceRect.size.width, FACE_RECT_SMOOTHING);
_smoothedFaceRect.size.height = glm::mix(faceRect.size.height, _smoothedFaceRect.size.height, FACE_RECT_SMOOTHING);
_smoothedFaceRect.angle = glm::mix(faceRect.angle, _smoothedFaceRect.angle, FACE_RECT_SMOOTHING);
}
// use the face rect to compute the face transform, aspect ratio
Point2f sourcePoints[4];
_smoothedFaceRect.points(sourcePoints);
Point2f destPoints[] = { Point2f(0, encodedHeight), Point2f(0, 0), Point2f(encodedWidth, 0) };
faceTransform = getAffineTransform(sourcePoints, destPoints);
aspectRatio = _smoothedFaceRect.size.width / _smoothedFaceRect.size.height;
}
const ushort ELEVEN_BIT_MINIMUM = 0;
const uchar EIGHT_BIT_MIDPOINT = 128;
double depthOffset;
if (!depth.empty()) {
if (_videoSendMode == FACE_VIDEO) {
// warp the face depth without interpolation (because it will contain invalid zero values)
_faceDepth.create(encodedHeight, encodedWidth, CV_16UC1);
warpAffine(depth, _faceDepth, faceTransform, _faceDepth.size(), INTER_NEAREST);
} else {
_faceDepth = depth;
}
_smoothedFaceDepth.create(encodedHeight, encodedWidth, CV_16UC1);
// smooth the depth over time
const ushort ELEVEN_BIT_MAXIMUM = 2047;
const float DEPTH_SMOOTHING = 0.25f;
ushort* src = _faceDepth.ptr<ushort>();
ushort* dest = _smoothedFaceDepth.ptr<ushort>();
ushort minimumDepth = numeric_limits<ushort>::max();
for (int i = 0; i < encodedHeight; i++) {
for (int j = 0; j < encodedWidth; j++) {
ushort depth = *src++;
if (depth != ELEVEN_BIT_MINIMUM && depth != ELEVEN_BIT_MAXIMUM) {
minimumDepth = min(minimumDepth, depth);
*dest = (*dest == ELEVEN_BIT_MINIMUM) ? depth : (ushort)glm::mix(depth, *dest, DEPTH_SMOOTHING);
}
dest++;
}
}
const ushort MINIMUM_DEPTH_OFFSET = 64;
const float FIXED_MID_DEPTH = 640.0f;
float midFaceDepth = (_videoSendMode == FACE_VIDEO) ? (minimumDepth + MINIMUM_DEPTH_OFFSET) : FIXED_MID_DEPTH;
// smooth the mid face depth over time
const float MID_FACE_DEPTH_SMOOTHING = 0.5f;
_smoothedMidFaceDepth = (_smoothedMidFaceDepth == UNINITIALIZED_FACE_DEPTH) ? midFaceDepth :
glm::mix(midFaceDepth, _smoothedMidFaceDepth, MID_FACE_DEPTH_SMOOTHING);
// convert from 11 to 8 bits for preview/local display // convert from 11 to 8 bits for preview/local display
const uchar EIGHT_BIT_MIDPOINT = 128; depthOffset = EIGHT_BIT_MIDPOINT - _smoothedMidFaceDepth;
double depthOffset = EIGHT_BIT_MIDPOINT - _smoothedMeanFaceDepth;
depth.convertTo(_grayDepthFrame, CV_8UC1, 1.0, depthOffset); depth.convertTo(_grayDepthFrame, CV_8UC1, 1.0, depthOffset);
}
// likewise for the encoded representation
QByteArray payload;
if (_videoSendMode != NO_VIDEO) {
if (_colorCodec.name == 0) {
// initialize encoder context(s)
vpx_codec_enc_cfg_t codecConfig;
vpx_codec_enc_config_default(vpx_codec_vp8_cx(), &codecConfig, 0);
codecConfig.rc_target_bitrate = encodedWidth * encodedHeight *
codecConfig.rc_target_bitrate / codecConfig.g_w / codecConfig.g_h;
codecConfig.g_w = encodedWidth;
codecConfig.g_h = encodedHeight;
vpx_codec_enc_init(&_colorCodec, vpx_codec_vp8_cx(), &codecConfig, 0);
if (!depth.empty()) {
codecConfig.rc_target_bitrate *= depthBitrateMultiplier;
vpx_codec_enc_init(&_depthCodec, vpx_codec_vp8_cx(), &codecConfig, 0);
}
}
Mat transform;
if (_videoSendMode == FACE_VIDEO) {
// resize/rotate face into encoding rectangle
_faceColor.create(encodedHeight, encodedWidth, CV_8UC3);
warpAffine(color, _faceColor, faceTransform, _faceColor.size());
} else {
_faceColor = color;
}
// convert from RGB to YV12: see http://www.fourcc.org/yuv.php and
// http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html#cvtcolor
const int ENCODED_BITS_PER_Y = 8;
const int ENCODED_BITS_PER_VU = 2;
const int ENCODED_BITS_PER_PIXEL = ENCODED_BITS_PER_Y + 2 * ENCODED_BITS_PER_VU;
const int BITS_PER_BYTE = 8;
_encodedFace.resize(encodedWidth * encodedHeight * ENCODED_BITS_PER_PIXEL / BITS_PER_BYTE);
vpx_image_t vpxImage;
vpx_img_wrap(&vpxImage, VPX_IMG_FMT_YV12, encodedWidth, encodedHeight, 1,
(unsigned char*)_encodedFace.data());
uchar* yline = vpxImage.planes[0]; uchar* yline = vpxImage.planes[0];
uchar* vline = vpxImage.planes[1]; uchar* vline = vpxImage.planes[1];
uchar* uline = vpxImage.planes[2]; uchar* uline = vpxImage.planes[2];
const uchar EIGHT_BIT_MAXIMUM = 255; const int Y_RED_WEIGHT = (int)(0.299 * 256);
for (int i = 0; i < ENCODED_FACE_HEIGHT; i += 2) { const int Y_GREEN_WEIGHT = (int)(0.587 * 256);
const int Y_BLUE_WEIGHT = (int)(0.114 * 256);
const int V_RED_WEIGHT = (int)(0.713 * 256);
const int U_BLUE_WEIGHT = (int)(0.564 * 256);
int redIndex = 0;
int greenIndex = 1;
int blueIndex = 2;
if (format == GL_BGR) {
redIndex = 2;
blueIndex = 0;
}
for (int i = 0; i < encodedHeight; i += 2) {
uchar* ydest = yline; uchar* ydest = yline;
uchar* vdest = vline; uchar* vdest = vline;
uchar* udest = uline; uchar* udest = uline;
for (int j = 0; j < ENCODED_FACE_WIDTH; j += 2) { for (int j = 0; j < encodedWidth; j += 2) {
ushort tl = *_faceDepth.ptr<ushort>(i, j); uchar* tl = _faceColor.ptr(i, j);
ushort tr = *_faceDepth.ptr<ushort>(i, j + 1); uchar* tr = _faceColor.ptr(i, j + 1);
ushort bl = *_faceDepth.ptr<ushort>(i + 1, j); uchar* bl = _faceColor.ptr(i + 1, j);
ushort br = *_faceDepth.ptr<ushort>(i + 1, j + 1); uchar* br = _faceColor.ptr(i + 1, j + 1);
uchar mask = EIGHT_BIT_MAXIMUM;
ydest[0] = (tl == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) : saturate_cast<uchar>(tl + depthOffset); ydest[0] = (tl[redIndex] * Y_RED_WEIGHT + tl[1] * Y_GREEN_WEIGHT + tl[blueIndex] * Y_BLUE_WEIGHT) >> 8;
ydest[1] = (tr == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) : saturate_cast<uchar>(tr + depthOffset); ydest[1] = (tr[redIndex] * Y_RED_WEIGHT + tr[1] * Y_GREEN_WEIGHT + tr[blueIndex] * Y_BLUE_WEIGHT) >> 8;
ydest[vpxImage.stride[0]] = (bl == ELEVEN_BIT_MINIMUM) ? ydest[vpxImage.stride[0]] = (bl[redIndex] * Y_RED_WEIGHT + bl[greenIndex] *
(mask = EIGHT_BIT_MIDPOINT) : saturate_cast<uchar>(bl + depthOffset); Y_GREEN_WEIGHT + bl[blueIndex] * Y_BLUE_WEIGHT) >> 8;
ydest[vpxImage.stride[0] + 1] = (br == ELEVEN_BIT_MINIMUM) ? ydest[vpxImage.stride[0] + 1] = (br[redIndex] * Y_RED_WEIGHT + br[greenIndex] *
(mask = EIGHT_BIT_MIDPOINT) : saturate_cast<uchar>(br + depthOffset); Y_GREEN_WEIGHT + br[blueIndex] * Y_BLUE_WEIGHT) >> 8;
ydest += 2; ydest += 2;
*vdest++ = mask; int totalRed = tl[redIndex] + tr[redIndex] + bl[redIndex] + br[redIndex];
*udest++ = EIGHT_BIT_MIDPOINT; int totalGreen = tl[greenIndex] + tr[greenIndex] + bl[greenIndex] + br[greenIndex];
int totalBlue = tl[blueIndex] + tr[blueIndex] + bl[blueIndex] + br[blueIndex];
int totalY = (totalRed * Y_RED_WEIGHT + totalGreen * Y_GREEN_WEIGHT + totalBlue * Y_BLUE_WEIGHT) >> 8;
*vdest++ = (((totalRed - totalY) * V_RED_WEIGHT) >> 10) + 128;
*udest++ = (((totalBlue - totalY) * U_BLUE_WEIGHT) >> 10) + 128;
} }
yline += vpxImage.stride[0] * 2; yline += vpxImage.stride[0] * 2;
vline += vpxImage.stride[1]; vline += vpxImage.stride[1];
@ -681,24 +674,79 @@ void FrameGrabber::grabFrame() {
} }
// encode the frame // encode the frame
vpx_codec_encode(&_depthCodec, &vpxImage, _frameCount, 1, 0, VPX_DL_REALTIME); vpx_codec_encode(&_colorCodec, &vpxImage, ++_frameCount, 1, 0, VPX_DL_REALTIME);
// start the payload off with the aspect ratio (zero for full frame)
payload.append((const char*)&aspectRatio, sizeof(float));
// extract the encoded frame // extract the encoded frame
vpx_codec_iter_t iterator = 0; vpx_codec_iter_t iterator = 0;
const vpx_codec_cx_pkt_t* packet; const vpx_codec_cx_pkt_t* packet;
while ((packet = vpx_codec_get_cx_data(&_depthCodec, &iterator)) != 0) { while ((packet = vpx_codec_get_cx_data(&_colorCodec, &iterator)) != 0) {
if (packet->kind == VPX_CODEC_CX_FRAME_PKT) { if (packet->kind == VPX_CODEC_CX_FRAME_PKT) {
// prepend the length, which will indicate whether there's a depth frame too
payload.append((const char*)&packet->data.frame.sz, sizeof(packet->data.frame.sz));
payload.append((const char*)packet->data.frame.buf, packet->data.frame.sz); payload.append((const char*)packet->data.frame.buf, packet->data.frame.sz);
} }
} }
if (!depth.empty()) {
// convert with mask
uchar* yline = vpxImage.planes[0];
uchar* vline = vpxImage.planes[1];
uchar* uline = vpxImage.planes[2];
const uchar EIGHT_BIT_MAXIMUM = 255;
for (int i = 0; i < encodedHeight; i += 2) {
uchar* ydest = yline;
uchar* vdest = vline;
uchar* udest = uline;
for (int j = 0; j < encodedWidth; j += 2) {
ushort tl = *_smoothedFaceDepth.ptr<ushort>(i, j);
ushort tr = *_smoothedFaceDepth.ptr<ushort>(i, j + 1);
ushort bl = *_smoothedFaceDepth.ptr<ushort>(i + 1, j);
ushort br = *_smoothedFaceDepth.ptr<ushort>(i + 1, j + 1);
uchar mask = EIGHT_BIT_MAXIMUM;
ydest[0] = (tl == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) :
saturate_cast<uchar>(tl + depthOffset);
ydest[1] = (tr == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) :
saturate_cast<uchar>(tr + depthOffset);
ydest[vpxImage.stride[0]] = (bl == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) :
saturate_cast<uchar>(bl + depthOffset);
ydest[vpxImage.stride[0] + 1] = (br == ELEVEN_BIT_MINIMUM) ? (mask = EIGHT_BIT_MIDPOINT) :
saturate_cast<uchar>(br + depthOffset);
ydest += 2;
*vdest++ = mask;
*udest++ = EIGHT_BIT_MIDPOINT;
}
yline += vpxImage.stride[0] * 2;
vline += vpxImage.stride[1];
uline += vpxImage.stride[2];
}
// encode the frame
vpx_codec_encode(&_depthCodec, &vpxImage, _frameCount, 1, 0, VPX_DL_REALTIME);
// extract the encoded frame
vpx_codec_iter_t iterator = 0;
const vpx_codec_cx_pkt_t* packet;
while ((packet = vpx_codec_get_cx_data(&_depthCodec, &iterator)) != 0) {
if (packet->kind == VPX_CODEC_CX_FRAME_PKT) {
payload.append((const char*)packet->data.frame.buf, packet->data.frame.sz);
}
}
}
} }
QMetaObject::invokeMethod(Application::getInstance(), "sendAvatarFaceVideoMessage", QMetaObject::invokeMethod(Application::getInstance(), "sendAvatarFaceVideoMessage",
Q_ARG(int, _frameCount), Q_ARG(QByteArray, payload)); Q_ARG(int, _frameCount), Q_ARG(QByteArray, payload));
QMetaObject::invokeMethod(Application::getInstance()->getWebcam(), "setFrame", QMetaObject::invokeMethod(Application::getInstance()->getWebcam(), "setFrame",
Q_ARG(cv::Mat, color), Q_ARG(int, format), Q_ARG(cv::Mat, _grayDepthFrame), Q_ARG(float, _smoothedMeanFaceDepth), Q_ARG(cv::Mat, color), Q_ARG(int, format), Q_ARG(cv::Mat, _grayDepthFrame), Q_ARG(float, _smoothedMidFaceDepth),
Q_ARG(cv::RotatedRect, _smoothedFaceRect), Q_ARG(JointVector, joints)); Q_ARG(float, aspectRatio), Q_ARG(cv::RotatedRect, _smoothedFaceRect), Q_ARG(bool, !payload.isEmpty()),
Q_ARG(JointVector, joints));
} }
bool FrameGrabber::init() { bool FrameGrabber::init() {
@ -767,6 +815,17 @@ void FrameGrabber::updateHSVFrame(const Mat& frame, int format) {
inRange(_hsvFrame, Scalar(0, 55, 65), Scalar(180, 256, 256), _mask); inRange(_hsvFrame, Scalar(0, 55, 65), Scalar(180, 256, 256), _mask);
} }
void FrameGrabber::destroyCodecs() {
if (_colorCodec.name != 0) {
vpx_codec_destroy(&_colorCodec);
_colorCodec.name = 0;
}
if (_depthCodec.name != 0) {
vpx_codec_destroy(&_depthCodec);
_depthCodec.name = 0;
}
}
Joint::Joint(const glm::vec3& position, const glm::quat& rotation, const glm::vec3& projected) : Joint::Joint(const glm::vec3& position, const glm::quat& rotation, const glm::vec3& projected) :
isValid(true), position(position), rotation(rotation), projected(projected) { isValid(true), position(position), rotation(rotation), projected(projected) {
} }

19
interface/src/Webcam.h
View file

@ -44,12 +44,17 @@ public:
Webcam(); Webcam();
~Webcam(); ~Webcam();
FrameGrabber* getGrabber() { return _grabber; }
bool isActive() const { return _active; } bool isActive() const { return _active; }
bool isSending() const { return _sending; }
GLuint getColorTextureID() const { return _colorTextureID; } GLuint getColorTextureID() const { return _colorTextureID; }
GLuint getDepthTextureID() const { return _depthTextureID; } GLuint getDepthTextureID() const { return _depthTextureID; }
const cv::Size2f& getTextureSize() const { return _textureSize; } const cv::Size2f& getTextureSize() const { return _textureSize; }
float getAspectRatio() const { return _aspectRatio; }
const cv::RotatedRect& getFaceRect() const { return _faceRect; } const cv::RotatedRect& getFaceRect() const { return _faceRect; }
const glm::vec3& getEstimatedPosition() const { return _estimatedPosition; } const glm::vec3& getEstimatedPosition() const { return _estimatedPosition; }
@ -62,8 +67,8 @@ public:
public slots: public slots:
void setEnabled(bool enabled); void setEnabled(bool enabled);
void setFrame(const cv::Mat& color, int format, const cv::Mat& depth, float meanFaceDepth, void setFrame(const cv::Mat& color, int format, const cv::Mat& depth, float midFaceDepth,
const cv::RotatedRect& faceRect, const JointVector& joints); float aspectRatio, const cv::RotatedRect& faceRect, bool sending, const JointVector& joints);
private: private:
@ -72,9 +77,11 @@ private:
bool _enabled; bool _enabled;
bool _active; bool _active;
bool _sending;
GLuint _colorTextureID; GLuint _colorTextureID;
GLuint _depthTextureID; GLuint _depthTextureID;
cv::Size2f _textureSize; cv::Size2f _textureSize;
float _aspectRatio;
cv::RotatedRect _faceRect; cv::RotatedRect _faceRect;
cv::RotatedRect _initialFaceRect; cv::RotatedRect _initialFaceRect;
float _initialFaceDepth; float _initialFaceDepth;
@ -100,16 +107,21 @@ public:
public slots: public slots:
void cycleVideoSendMode();
void reset(); void reset();
void shutdown(); void shutdown();
void grabFrame(); void grabFrame();
private: private:
enum VideoSendMode { NO_VIDEO, FACE_VIDEO, FULL_FRAME_VIDEO, VIDEO_SEND_MODE_COUNT };
bool init(); bool init();
void updateHSVFrame(const cv::Mat& frame, int format); void updateHSVFrame(const cv::Mat& frame, int format);
void destroyCodecs();
bool _initialized; bool _initialized;
VideoSendMode _videoSendMode;
CvCapture* _capture; CvCapture* _capture;
cv::CascadeClassifier _faceCascade; cv::CascadeClassifier _faceCascade;
cv::Mat _hsvFrame; cv::Mat _hsvFrame;
@ -118,13 +130,14 @@ private:
cv::Mat _backProject; cv::Mat _backProject;
cv::Rect _searchWindow; cv::Rect _searchWindow;
cv::Mat _grayDepthFrame; cv::Mat _grayDepthFrame;
float _smoothedMeanFaceDepth; float _smoothedMidFaceDepth;
vpx_codec_ctx_t _colorCodec; vpx_codec_ctx_t _colorCodec;
vpx_codec_ctx_t _depthCodec; vpx_codec_ctx_t _depthCodec;
int _frameCount; int _frameCount;
cv::Mat _faceColor; cv::Mat _faceColor;
cv::Mat _faceDepth; cv::Mat _faceDepth;
cv::Mat _smoothedFaceDepth;
QByteArray _encodedFace; QByteArray _encodedFace;
cv::RotatedRect _smoothedFaceRect; cv::RotatedRect _smoothedFaceRect;

19
interface/src/avatar/Avatar.cpp
View file

@ -314,10 +314,7 @@ void Avatar::updateFromGyrosAndOrWebcam(bool gyroLook,
estimatedPosition = webcam->getEstimatedPosition(); estimatedPosition = webcam->getEstimatedPosition();
// apply face data // apply face data
_head.getFace().setColorTextureID(webcam->getColorTextureID()); _head.getFace().setFrameFromWebcam();
_head.getFace().setDepthTextureID(webcam->getDepthTextureID());
_head.getFace().setTextureSize(webcam->getTextureSize());
_head.getFace().setTextureRect(webcam->getFaceRect());
// compute and store the joint rotations // compute and store the joint rotations
const JointVector& joints = webcam->getEstimatedJoints(); const JointVector& joints = webcam->getEstimatedJoints();
@ -334,7 +331,7 @@ void Avatar::updateFromGyrosAndOrWebcam(bool gyroLook,
} }
} }
} else { } else {
_head.getFace().setColorTextureID(0); _head.getFace().clearFrame();
} }
_head.setPitch(estimatedRotation.x * amplifyAngle.x + pitchFromTouch); _head.setPitch(estimatedRotation.x * amplifyAngle.x + pitchFromTouch);
_head.setYaw(estimatedRotation.y * amplifyAngle.y + yawFromTouch); _head.setYaw(estimatedRotation.y * amplifyAngle.y + yawFromTouch);
@ -1300,9 +1297,15 @@ float Avatar::getBallRenderAlpha(int ball, bool lookingInMirror) const {
} }
void Avatar::renderBody(bool lookingInMirror, bool renderAvatarBalls) { void Avatar::renderBody(bool lookingInMirror, bool renderAvatarBalls) {
// Render the body as balls and cones if (_head.getFace().isFullFrame()) {
if (renderAvatarBalls || !_voxels.getVoxelURL().isValid()) { // Render the full-frame video
float alpha = getBallRenderAlpha(BODY_BALL_HEAD_BASE, lookingInMirror);
if (alpha > 0.0f) {
_head.getFace().render(1.0f);
}
} else if (renderAvatarBalls || !_voxels.getVoxelURL().isValid()) {
// Render the body as balls and cones
for (int b = 0; b < NUM_AVATAR_BODY_BALLS; b++) { for (int b = 0; b < NUM_AVATAR_BODY_BALLS; b++) {
float alpha = getBallRenderAlpha(b, lookingInMirror); float alpha = getBallRenderAlpha(b, lookingInMirror);

319
interface/src/avatar/Face.cpp
View file

@ -17,6 +17,7 @@
#include "Avatar.h" #include "Avatar.h"
#include "Head.h" #include "Head.h"
#include "Face.h" #include "Face.h"
#include "Webcam.h"
#include "renderer/ProgramObject.h" #include "renderer/ProgramObject.h"
using namespace cv; using namespace cv;
@ -25,7 +26,6 @@ ProgramObject* Face::_program = 0;
int Face::_texCoordCornerLocation; int Face::_texCoordCornerLocation;
int Face::_texCoordRightLocation; int Face::_texCoordRightLocation;
int Face::_texCoordUpLocation; int Face::_texCoordUpLocation;
int Face::_aspectRatioLocation;
GLuint Face::_vboID; GLuint Face::_vboID;
GLuint Face::_iboID; GLuint Face::_iboID;
@ -55,17 +55,25 @@ Face::~Face() {
} }
} }
void Face::setTextureRect(const cv::RotatedRect& textureRect) { void Face::setFrameFromWebcam() {
_textureRect = textureRect; Webcam* webcam = Application::getInstance()->getWebcam();
_aspectRatio = _textureRect.size.width / _textureRect.size.height; if (webcam->isSending()) {
_colorTextureID = webcam->getColorTextureID();
_depthTextureID = webcam->getDepthTextureID();
_textureSize = webcam->getTextureSize();
_textureRect = webcam->getFaceRect();
_aspectRatio = webcam->getAspectRatio();
} else {
clearFrame();
}
} }
void Face::clearFrame() {
_colorTextureID = 0;
}
int Face::processVideoMessage(unsigned char* packetData, size_t dataBytes) { int Face::processVideoMessage(unsigned char* packetData, size_t dataBytes) {
if (_colorCodec.name == 0) {
// initialize decoder context
vpx_codec_dec_init(&_colorCodec, vpx_codec_vp8_dx(), 0, 0);
}
// skip the header
unsigned char* packetPosition = packetData; unsigned char* packetPosition = packetData;
int frameCount = *(uint32_t*)packetPosition; int frameCount = *(uint32_t*)packetPosition;
@ -89,110 +97,135 @@ int Face::processVideoMessage(unsigned char* packetData, size_t dataBytes) {
int payloadSize = dataBytes - (packetPosition - packetData); int payloadSize = dataBytes - (packetPosition - packetData);
memcpy(_arrivingFrame.data() + frameOffset, packetPosition, payloadSize); memcpy(_arrivingFrame.data() + frameOffset, packetPosition, payloadSize);
if ((_frameBytesRemaining -= payloadSize) <= 0) { if ((_frameBytesRemaining -= payloadSize) > 0) {
float aspectRatio = *(const float*)_arrivingFrame.constData(); return dataBytes; // wait for the rest of the frame
size_t colorSize = *(const size_t*)(_arrivingFrame.constData() + sizeof(float)); }
const uint8_t* colorData = (const uint8_t*)(_arrivingFrame.constData() + sizeof(float) + sizeof(size_t));
vpx_codec_decode(&_colorCodec, colorData, colorSize, 0, 0); if (frameSize == 0) {
vpx_codec_iter_t iterator = 0; // destroy the codecs, if we have any
vpx_image_t* image; destroyCodecs();
while ((image = vpx_codec_get_frame(&_colorCodec, &iterator)) != 0) {
// convert from YV12 to RGB // disables video data
Mat color(image->d_h, image->d_w, CV_8UC3); QMetaObject::invokeMethod(this, "setFrame", Q_ARG(cv::Mat, Mat()),
uchar* yline = image->planes[0]; Q_ARG(cv::Mat, Mat()), Q_ARG(float, 0.0f));
uchar* vline = image->planes[1]; return dataBytes;
uchar* uline = image->planes[2]; }
const int RED_V_WEIGHT = (int)(1.403 * 256);
const int GREEN_V_WEIGHT = (int)(0.714 * 256); // the switch from full frame to not (or vice versa) requires us to reinit the codecs
const int GREEN_U_WEIGHT = (int)(0.344 * 256); float aspectRatio = *(const float*)_arrivingFrame.constData();
const int BLUE_U_WEIGHT = (int)(1.773 * 256); bool fullFrame = (aspectRatio == FULL_FRAME_ASPECT);
for (int i = 0; i < image->d_h; i += 2) { if (fullFrame != _lastFullFrame) {
uchar* ysrc = yline; destroyCodecs();
uchar* vsrc = vline; _lastFullFrame = fullFrame;
uchar* usrc = uline; }
for (int j = 0; j < image->d_w; j += 2) {
uchar* tl = color.ptr(i, j); if (_colorCodec.name == 0) {
uchar* tr = color.ptr(i, j + 1); // initialize decoder context
uchar* bl = color.ptr(i + 1, j); vpx_codec_dec_init(&_colorCodec, vpx_codec_vp8_dx(), 0, 0);
uchar* br = color.ptr(i + 1, j + 1); }
int v = *vsrc++ - 128; size_t colorSize = *(const size_t*)(_arrivingFrame.constData() + sizeof(float));
int u = *usrc++ - 128; const uint8_t* colorData = (const uint8_t*)(_arrivingFrame.constData() + sizeof(float) + sizeof(size_t));
vpx_codec_decode(&_colorCodec, colorData, colorSize, 0, 0);
int redOffset = (RED_V_WEIGHT * v) >> 8; vpx_codec_iter_t iterator = 0;
int greenOffset = (GREEN_V_WEIGHT * v + GREEN_U_WEIGHT * u) >> 8; vpx_image_t* image;
int blueOffset = (BLUE_U_WEIGHT * u) >> 8; while ((image = vpx_codec_get_frame(&_colorCodec, &iterator)) != 0) {
// convert from YV12 to RGB: see http://www.fourcc.org/yuv.php and
int ytl = ysrc[0]; // http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html#cvtcolor
int ytr = ysrc[1]; Mat color(image->d_h, image->d_w, CV_8UC3);
int ybl = ysrc[image->w]; uchar* yline = image->planes[0];
int ybr = ysrc[image->w + 1]; uchar* vline = image->planes[1];
ysrc += 2; uchar* uline = image->planes[2];
const int RED_V_WEIGHT = (int)(1.403 * 256);
tl[0] = ytl + redOffset; const int GREEN_V_WEIGHT = (int)(0.714 * 256);
tl[1] = ytl - greenOffset; const int GREEN_U_WEIGHT = (int)(0.344 * 256);
tl[2] = ytl + blueOffset; const int BLUE_U_WEIGHT = (int)(1.773 * 256);
for (int i = 0; i < image->d_h; i += 2) {
tr[0] = ytr + redOffset; uchar* ysrc = yline;
tr[1] = ytr - greenOffset; uchar* vsrc = vline;
tr[2] = ytr + blueOffset; uchar* usrc = uline;
for (int j = 0; j < image->d_w; j += 2) {
bl[0] = ybl + redOffset; uchar* tl = color.ptr(i, j);
bl[1] = ybl - greenOffset; uchar* tr = color.ptr(i, j + 1);
bl[2] = ybl + blueOffset; uchar* bl = color.ptr(i + 1, j);
uchar* br = color.ptr(i + 1, j + 1);
br[0] = ybr + redOffset;
br[1] = ybr - greenOffset; int v = *vsrc++ - 128;
br[2] = ybr + blueOffset; int u = *usrc++ - 128;
}
yline += image->stride[0] * 2; int redOffset = (RED_V_WEIGHT * v) >> 8;
vline += image->stride[1]; int greenOffset = (GREEN_V_WEIGHT * v + GREEN_U_WEIGHT * u) >> 8;
uline += image->stride[2]; int blueOffset = (BLUE_U_WEIGHT * u) >> 8;
int ytl = ysrc[0];
int ytr = ysrc[1];
int ybl = ysrc[image->w];
int ybr = ysrc[image->w + 1];
ysrc += 2;
tl[0] = ytl + redOffset;
tl[1] = ytl - greenOffset;
tl[2] = ytl + blueOffset;
tr[0] = ytr + redOffset;
tr[1] = ytr - greenOffset;
tr[2] = ytr + blueOffset;
bl[0] = ybl + redOffset;
bl[1] = ybl - greenOffset;
bl[2] = ybl + blueOffset;
br[0] = ybr + redOffset;
br[1] = ybr - greenOffset;
br[2] = ybr + blueOffset;
} }
Mat depth; yline += image->stride[0] * 2;
vline += image->stride[1];
const uint8_t* depthData = colorData + colorSize; uline += image->stride[2];
int depthSize = _arrivingFrame.size() - ((const char*)depthData - _arrivingFrame.constData());
if (depthSize > 0) {
if (_depthCodec.name == 0) {
// initialize decoder context
vpx_codec_dec_init(&_depthCodec, vpx_codec_vp8_dx(), 0, 0);
}
vpx_codec_decode(&_depthCodec, depthData, depthSize, 0, 0);
vpx_codec_iter_t iterator = 0;
vpx_image_t* image;
while ((image = vpx_codec_get_frame(&_depthCodec, &iterator)) != 0) {
depth.create(image->d_h, image->d_w, CV_8UC1);
uchar* yline = image->planes[0];
uchar* vline = image->planes[1];
const uchar EIGHT_BIT_MAXIMUM = 255;
const uchar MASK_THRESHOLD = 192;
for (int i = 0; i < image->d_h; i += 2) {
uchar* ysrc = yline;
uchar* vsrc = vline;
for (int j = 0; j < image->d_w; j += 2) {
if (*vsrc++ < MASK_THRESHOLD) {
*depth.ptr(i, j) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i, j + 1) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i + 1, j) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i + 1, j + 1) = EIGHT_BIT_MAXIMUM;
} else {
*depth.ptr(i, j) = ysrc[0];
*depth.ptr(i, j + 1) = ysrc[1];
*depth.ptr(i + 1, j) = ysrc[image->stride[0]];
*depth.ptr(i + 1, j + 1) = ysrc[image->stride[0] + 1];
}
ysrc += 2;
}
yline += image->stride[0] * 2;
vline += image->stride[1];
}
}
}
QMetaObject::invokeMethod(this, "setFrame", Q_ARG(cv::Mat, color),
Q_ARG(cv::Mat, depth), Q_ARG(float, aspectRatio));
} }
Mat depth;
const uint8_t* depthData = colorData + colorSize;
int depthSize = _arrivingFrame.size() - ((const char*)depthData - _arrivingFrame.constData());
if (depthSize > 0) {
if (_depthCodec.name == 0) {
// initialize decoder context
vpx_codec_dec_init(&_depthCodec, vpx_codec_vp8_dx(), 0, 0);
}
vpx_codec_decode(&_depthCodec, depthData, depthSize, 0, 0);
vpx_codec_iter_t iterator = 0;
vpx_image_t* image;
while ((image = vpx_codec_get_frame(&_depthCodec, &iterator)) != 0) {
depth.create(image->d_h, image->d_w, CV_8UC1);
uchar* yline = image->planes[0];
uchar* vline = image->planes[1];
const uchar EIGHT_BIT_MAXIMUM = 255;
const uchar MASK_THRESHOLD = 192;
for (int i = 0; i < image->d_h; i += 2) {
uchar* ysrc = yline;
uchar* vsrc = vline;
for (int j = 0; j < image->d_w; j += 2) {
if (*vsrc++ < MASK_THRESHOLD) {
*depth.ptr(i, j) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i, j + 1) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i + 1, j) = EIGHT_BIT_MAXIMUM;
*depth.ptr(i + 1, j + 1) = EIGHT_BIT_MAXIMUM;
} else {
*depth.ptr(i, j) = ysrc[0];
*depth.ptr(i, j + 1) = ysrc[1];
*depth.ptr(i + 1, j) = ysrc[image->stride[0]];
*depth.ptr(i + 1, j + 1) = ysrc[image->stride[0] + 1];
}
ysrc += 2;
}
yline += image->stride[0] * 2;
vline += image->stride[1];
}
}
}
QMetaObject::invokeMethod(this, "setFrame", Q_ARG(cv::Mat, color),
Q_ARG(cv::Mat, depth), Q_ARG(float, aspectRatio));
} }
return dataBytes; return dataBytes;
@ -208,9 +241,22 @@ bool Face::render(float alpha) {
glm::quat orientation = _owningHead->getOrientation(); glm::quat orientation = _owningHead->getOrientation();
glm::vec3 axis = glm::axis(orientation); glm::vec3 axis = glm::axis(orientation);
glRotatef(glm::angle(orientation), axis.x, axis.y, axis.z); glRotatef(glm::angle(orientation), axis.x, axis.y, axis.z);
float scale = BODY_BALL_RADIUS_HEAD_BASE * _owningHead->getScale();
glScalef(scale, scale, scale); float aspect, xScale, zScale;
if (_aspectRatio == FULL_FRAME_ASPECT) {
aspect = _textureSize.width / _textureSize.height;
const float FULL_FRAME_SCALE = 0.5f;
xScale = FULL_FRAME_SCALE * _owningHead->getScale();
zScale = xScale * 0.3f;
} else {
aspect = _aspectRatio;
xScale = BODY_BALL_RADIUS_HEAD_BASE * _owningHead->getScale();
zScale = xScale * 1.5f;
glTranslatef(0.0f, -xScale * 0.75f, -xScale);
}
glScalef(xScale, xScale / aspect, zScale);
glColor4f(1.0f, 1.0f, 1.0f, alpha); glColor4f(1.0f, 1.0f, 1.0f, alpha);
Point2f points[4]; Point2f points[4];
@ -243,7 +289,6 @@ bool Face::render(float alpha) {
_texCoordCornerLocation = _program->uniformLocation("texCoordCorner"); _texCoordCornerLocation = _program->uniformLocation("texCoordCorner");
_texCoordRightLocation = _program->uniformLocation("texCoordRight"); _texCoordRightLocation = _program->uniformLocation("texCoordRight");
_texCoordUpLocation = _program->uniformLocation("texCoordUp"); _texCoordUpLocation = _program->uniformLocation("texCoordUp");
_aspectRatioLocation = _program->uniformLocation("aspectRatio");
glGenBuffers(1, &_vboID); glGenBuffers(1, &_vboID);
glBindBuffer(GL_ARRAY_BUFFER, _vboID); glBindBuffer(GL_ARRAY_BUFFER, _vboID);
@ -292,7 +337,6 @@ bool Face::render(float alpha) {
(points[3].x - points[0].x) / _textureSize.width, (points[3].y - points[0].y) / _textureSize.height); (points[3].x - points[0].x) / _textureSize.width, (points[3].y - points[0].y) / _textureSize.height);
_program->setUniformValue(_texCoordUpLocation, _program->setUniformValue(_texCoordUpLocation,
(points[1].x - points[0].x) / _textureSize.width, (points[1].y - points[0].y) / _textureSize.height); (points[1].x - points[0].x) / _textureSize.width, (points[1].y - points[0].y) / _textureSize.height);
_program->setUniformValue(_aspectRatioLocation, _aspectRatio);
glEnableClientState(GL_VERTEX_ARRAY); glEnableClientState(GL_VERTEX_ARRAY);
glVertexPointer(2, GL_FLOAT, 0, 0); glVertexPointer(2, GL_FLOAT, 0, 0);
@ -324,13 +368,13 @@ bool Face::render(float alpha) {
glBegin(GL_QUADS); glBegin(GL_QUADS);
glTexCoord2f(points[0].x / _textureSize.width, points[0].y / _textureSize.height); glTexCoord2f(points[0].x / _textureSize.width, points[0].y / _textureSize.height);
glVertex3f(0.5f, -0.5f / _aspectRatio, -0.5f); glVertex3f(0.5f, -0.5f, 0.0f);
glTexCoord2f(points[1].x / _textureSize.width, points[1].y / _textureSize.height); glTexCoord2f(points[1].x / _textureSize.width, points[1].y / _textureSize.height);
glVertex3f(0.5f, 0.5f / _aspectRatio, -0.5f); glVertex3f(0.5f, 0.5f, 0.0f);
glTexCoord2f(points[2].x / _textureSize.width, points[2].y / _textureSize.height); glTexCoord2f(points[2].x / _textureSize.width, points[2].y / _textureSize.height);
glVertex3f(-0.5f, 0.5f / _aspectRatio, -0.5f); glVertex3f(-0.5f, 0.5f, 0.0f);
glTexCoord2f(points[3].x / _textureSize.width, points[3].y / _textureSize.height); glTexCoord2f(points[3].x / _textureSize.width, points[3].y / _textureSize.height);
glVertex3f(-0.5f, -0.5f / _aspectRatio, -0.5f); glVertex3f(-0.5f, -0.5f, 0.0f);
glEnd(); glEnd();
glDisable(GL_TEXTURE_2D); glDisable(GL_TEXTURE_2D);
@ -348,23 +392,40 @@ void Face::cycleRenderMode() {
} }
void Face::setFrame(const cv::Mat& color, const cv::Mat& depth, float aspectRatio) { void Face::setFrame(const cv::Mat& color, const cv::Mat& depth, float aspectRatio) {
if (color.empty()) {
// release our textures, if any; there's no more video
if (_colorTextureID != 0) {
glDeleteTextures(1, &_colorTextureID);
_colorTextureID = 0;
}
if (_depthTextureID != 0) {
glDeleteTextures(1, &_depthTextureID);
_depthTextureID = 0;
}
return;
}
if (_colorTextureID == 0) { if (_colorTextureID == 0) {
glGenTextures(1, &_colorTextureID); glGenTextures(1, &_colorTextureID);
glBindTexture(GL_TEXTURE_2D, _colorTextureID); }
glBindTexture(GL_TEXTURE_2D, _colorTextureID);
bool recreateTextures = (_textureSize.width != color.cols || _textureSize.height != color.rows);
if (recreateTextures) {
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB, color.cols, color.rows, 0, GL_RGB, GL_UNSIGNED_BYTE, color.ptr()); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB, color.cols, color.rows, 0, GL_RGB, GL_UNSIGNED_BYTE, color.ptr());
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
_textureSize = color.size(); _textureSize = color.size();
_textureRect = RotatedRect(Point2f(color.cols * 0.5f, color.rows * 0.5f), _textureSize, 0.0f); _textureRect = RotatedRect(Point2f(color.cols * 0.5f, color.rows * 0.5f), _textureSize, 0.0f);
} else { } else {
glBindTexture(GL_TEXTURE_2D, _colorTextureID);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, color.cols, color.rows, GL_RGB, GL_UNSIGNED_BYTE, color.ptr()); glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, color.cols, color.rows, GL_RGB, GL_UNSIGNED_BYTE, color.ptr());
} }
if (!depth.empty()) { if (!depth.empty()) {
if (_depthTextureID == 0) { if (_depthTextureID == 0) {
glGenTextures(1, &_depthTextureID); glGenTextures(1, &_depthTextureID);
glBindTexture(GL_TEXTURE_2D, _depthTextureID); }
glBindTexture(GL_TEXTURE_2D, _depthTextureID);
if (recreateTextures) {
glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, depth.cols, depth.rows, 0, glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, depth.cols, depth.rows, 0,
GL_LUMINANCE, GL_UNSIGNED_BYTE, depth.ptr()); GL_LUMINANCE, GL_UNSIGNED_BYTE, depth.ptr());
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
@ -380,3 +441,13 @@ void Face::setFrame(const cv::Mat& color, const cv::Mat& depth, float aspectRati
_aspectRatio = aspectRatio; _aspectRatio = aspectRatio;
} }
void Face::destroyCodecs() {
if (_colorCodec.name != 0) {
vpx_codec_destroy(&_colorCodec);
_colorCodec.name = 0;
}
if (_depthCodec.name != 0) {
vpx_codec_destroy(&_depthCodec);
_depthCodec.name = 0;
}
}

14
interface/src/avatar/Face.h
View file

@ -20,6 +20,8 @@
class Head; class Head;
class ProgramObject; class ProgramObject;
const float FULL_FRAME_ASPECT = 0.0f;
class Face : public QObject { class Face : public QObject {
Q_OBJECT Q_OBJECT
@ -28,10 +30,10 @@ public:
Face(Head* owningHead); Face(Head* owningHead);
~Face(); ~Face();
void setColorTextureID(GLuint colorTextureID) { _colorTextureID = colorTextureID; } bool isFullFrame() const { return _colorTextureID != 0 && _aspectRatio == FULL_FRAME_ASPECT; }
void setDepthTextureID(GLuint depthTextureID) { _depthTextureID = depthTextureID; }
void setTextureSize(const cv::Size2f& textureSize) { _textureSize = textureSize; } void setFrameFromWebcam();
void setTextureRect(const cv::RotatedRect& textureRect); void clearFrame();
int processVideoMessage(unsigned char* packetData, size_t dataBytes); int processVideoMessage(unsigned char* packetData, size_t dataBytes);
@ -49,6 +51,8 @@ private:
enum RenderMode { MESH, POINTS, RENDER_MODE_COUNT }; enum RenderMode { MESH, POINTS, RENDER_MODE_COUNT };
void destroyCodecs();
Head* _owningHead; Head* _owningHead;
RenderMode _renderMode; RenderMode _renderMode;
GLuint _colorTextureID; GLuint _colorTextureID;
@ -59,6 +63,7 @@ private:
vpx_codec_ctx_t _colorCodec; vpx_codec_ctx_t _colorCodec;
vpx_codec_ctx_t _depthCodec; vpx_codec_ctx_t _depthCodec;
bool _lastFullFrame;
QByteArray _arrivingFrame; QByteArray _arrivingFrame;
int _frameCount; int _frameCount;
@ -68,7 +73,6 @@ private:
static int _texCoordCornerLocation; static int _texCoordCornerLocation;
static int _texCoordRightLocation; static int _texCoordRightLocation;
static int _texCoordUpLocation; static int _texCoordUpLocation;
static int _aspectRatioLocation;
static GLuint _vboID; static GLuint _vboID;
static GLuint _iboID; static GLuint _iboID;
}; };