mirror of
https://github.com/introlab/rtabmap.git
synced 2026-09-02 09:30:25 +08:00
android: added occlusion detection using arcore depth api (for FPS view). Added depth from motion option in Mapping options.
This commit is contained in:
@@ -37,14 +37,16 @@ namespace rtabmap {
|
||||
//////////////////////////////
|
||||
// CameraARCore
|
||||
//////////////////////////////
|
||||
CameraARCore::CameraARCore(void* env, void* context, void* activity, bool smoothing):
|
||||
CameraARCore::CameraARCore(void* env, void* context, void* activity, bool depthFromMotion, bool smoothing):
|
||||
CameraMobile(smoothing),
|
||||
env_(env),
|
||||
context_(context),
|
||||
activity_(activity),
|
||||
arInstallRequested_(false),
|
||||
textureId_(9999),
|
||||
uvs_initialized_(false)
|
||||
uvs_initialized_(false),
|
||||
updateOcclusionImage_(false),
|
||||
depthFromMotion_(depthFromMotion)
|
||||
{
|
||||
}
|
||||
|
||||
@@ -162,8 +164,8 @@ bool CameraARCore::init(const std::string & calibrationFolder, const std::string
|
||||
UASSERT(ArSession_create(env_, context_, &arSession_) == AR_SUCCESS);
|
||||
UASSERT(arSession_);
|
||||
|
||||
int32_t is_depth_supported = 0; // Disabled by default, depth is not super accurate for mapping
|
||||
//ArSession_isDepthModeSupported(arSession_, AR_DEPTH_MODE_AUTOMATIC, &is_depth_supported);
|
||||
int32_t is_depth_supported = 0;
|
||||
ArSession_isDepthModeSupported(arSession_, AR_DEPTH_MODE_AUTOMATIC, &is_depth_supported);
|
||||
|
||||
ArConfig_create(arSession_, &arConfig_);
|
||||
UASSERT(arConfig_);
|
||||
@@ -277,6 +279,7 @@ void CameraARCore::close()
|
||||
arPose_ = nullptr;
|
||||
|
||||
CameraMobile::close();
|
||||
occlusionImage_ = cv::Mat();
|
||||
}
|
||||
|
||||
LaserScan CameraARCore::scanFromPointCloudData(
|
||||
@@ -429,8 +432,7 @@ SensorData CameraARCore::captureImage(CameraInfo * info)
|
||||
ArStatus status = ArFrame_acquireCameraImage(arSession_, arFrame_, &image);
|
||||
if(status == AR_SUCCESS)
|
||||
{
|
||||
cv::Mat outputDepth;
|
||||
if(is_depth_supported)
|
||||
if(is_depth_supported && (updateOcclusionImage_||depthFromMotion_))
|
||||
{
|
||||
LOGD("Acquire depth image!");
|
||||
ArImage * depthImage = nullptr;
|
||||
@@ -444,36 +446,24 @@ SensorData CameraARCore::captureImage(CameraInfo * info)
|
||||
int planeCount;
|
||||
ArImage_getNumberOfPlanes(arSession_, depthImage, &planeCount);
|
||||
LOGD("planeCount=%d", planeCount);
|
||||
UASSERT_MSG(planeCount == 1,
|
||||
uFormat("Error: getNumberOfPlanes() planceCount = %d", planeCount).c_str());
|
||||
UASSERT_MSG(planeCount == 1, uFormat("Error: getNumberOfPlanes() planceCount = %d", planeCount).c_str());
|
||||
const uint8_t *data = nullptr;
|
||||
int len = 0;
|
||||
int stride;
|
||||
int width;
|
||||
int height;
|
||||
ArImage_getWidth(arSession_, depthImage, &width);
|
||||
ArImage_getHeight(arSession_, depthImage, &height);
|
||||
int depth_width;
|
||||
int depth_height;
|
||||
ArImage_getWidth(arSession_, depthImage, &depth_width);
|
||||
ArImage_getHeight(arSession_, depthImage, &depth_height);
|
||||
ArImage_getPlaneRowStride(arSession_, depthImage, 0, &stride);
|
||||
ArImage_getPlaneData(arSession_, depthImage, 0, &data, &len);
|
||||
|
||||
LOGD("width=%d, height=%d, bytes=%d stride=%d", width, height, len, stride);
|
||||
LOGD("width=%d, height=%d, bytes=%d stride=%d", depth_width, depth_height, len, stride);
|
||||
|
||||
outputDepth = cv::Mat(height, width, CV_16UC1);
|
||||
uint16_t *dataShort = (uint16_t *)data;
|
||||
uint16_t max=0x0;
|
||||
for (int y = 0; y < outputDepth.rows; ++y)
|
||||
{
|
||||
for (int x = 0; x < outputDepth.cols; ++x)
|
||||
{
|
||||
uint16_t depthSample = dataShort[y*outputDepth.cols + x];
|
||||
uint16_t depthRange = (depthSample & 0x1FFF); // first 3 bits are confidence
|
||||
outputDepth.at<uint16_t>(y,x) = depthRange;
|
||||
if(depthRange > max)
|
||||
{
|
||||
max = depthRange;
|
||||
}
|
||||
}
|
||||
}
|
||||
occlusionImage_ = cv::Mat(depth_height, depth_width, CV_16UC1, (void*)data).clone();
|
||||
|
||||
float scaleX = (float)depth_width / (float)width;
|
||||
float scaleY = (float)depth_height / (float)height;
|
||||
occlusionModel_ = CameraModel(fx*scaleX, fy*scaleY, cx*scaleX, cy*scaleY, pose*deviceTColorCamera_, 0, cv::Size(depth_width, depth_height));
|
||||
}
|
||||
ArImage_release(depthImage);
|
||||
}
|
||||
@@ -545,7 +535,7 @@ SensorData CameraARCore::captureImage(CameraInfo * info)
|
||||
LOGI("pointCloud empty");
|
||||
}
|
||||
|
||||
data = SensorData(scan, rgb, outputDepth, model, 0, stamp);
|
||||
data = SensorData(scan, rgb, depthFromMotion_?occlusionImage_:cv::Mat(), model, 0, stamp);
|
||||
data.setFeatures(kpts, kpts3, cv::Mat());
|
||||
}
|
||||
}
|
||||
@@ -619,11 +609,6 @@ void CameraARCore::capturePoseOnly()
|
||||
uvs_initialized_ = true;
|
||||
}
|
||||
|
||||
/*ArImage* ar_image = nullptr;
|
||||
ArStatus status =
|
||||
ArFrame_acquireCameraImage(arSession_, arFrame_, &ar_image);
|
||||
ArImage_release(ar_image);
|
||||
*/
|
||||
ArCamera* ar_camera;
|
||||
ArFrame_acquireCamera(arSession_, arFrame_, &ar_camera);
|
||||
|
||||
@@ -649,6 +634,52 @@ void CameraARCore::capturePoseOnly()
|
||||
pose = rtabmap::rtabmap_world_T_opengl_world * pose * rtabmap::opengl_world_T_rtabmap_world;
|
||||
this->poseReceived(pose);
|
||||
}
|
||||
|
||||
int32_t is_depth_supported = 0;
|
||||
ArSession_isDepthModeSupported(arSession_, AR_DEPTH_MODE_AUTOMATIC, &is_depth_supported);
|
||||
|
||||
if(is_depth_supported && updateOcclusionImage_)
|
||||
{
|
||||
LOGD("Acquire depth image!");
|
||||
ArImage * depthImage = nullptr;
|
||||
ArFrame_acquireDepthImage(arSession_, arFrame_, &depthImage);
|
||||
|
||||
ArImageFormat format;
|
||||
ArImage_getFormat(arSession_, depthImage, &format);
|
||||
if(format == AR_IMAGE_FORMAT_DEPTH16)
|
||||
{
|
||||
LOGD("Depth format detected!");
|
||||
int planeCount;
|
||||
ArImage_getNumberOfPlanes(arSession_, depthImage, &planeCount);
|
||||
LOGD("planeCount=%d", planeCount);
|
||||
UASSERT_MSG(planeCount == 1, uFormat("Error: getNumberOfPlanes() planceCount = %d", planeCount).c_str());
|
||||
const uint8_t *data = nullptr;
|
||||
int len = 0;
|
||||
int stride;
|
||||
int width;
|
||||
int height;
|
||||
ArImage_getWidth(arSession_, depthImage, &width);
|
||||
ArImage_getHeight(arSession_, depthImage, &height);
|
||||
ArImage_getPlaneRowStride(arSession_, depthImage, 0, &stride);
|
||||
ArImage_getPlaneData(arSession_, depthImage, 0, &data, &len);
|
||||
|
||||
LOGD("width=%d, height=%d, bytes=%d stride=%d", width, height, len, stride);
|
||||
|
||||
occlusionImage_ = cv::Mat(height, width, CV_16UC1, (void*)data).clone();
|
||||
|
||||
float fx,fy, cx, cy;
|
||||
int32_t rgb_width, rgb_height;
|
||||
ArCamera_getImageIntrinsics(arSession_, ar_camera, arCameraIntrinsics_);
|
||||
ArCameraIntrinsics_getFocalLength(arSession_, arCameraIntrinsics_, &fx, &fy);
|
||||
ArCameraIntrinsics_getPrincipalPoint(arSession_, arCameraIntrinsics_, &cx, &cy);
|
||||
ArCameraIntrinsics_getImageDimensions(arSession_, arCameraIntrinsics_, &rgb_width, &rgb_height);
|
||||
|
||||
float scaleX = (float)width / (float)rgb_width;
|
||||
float scaleY = (float)height / (float)rgb_height;
|
||||
occlusionModel_ = CameraModel(fx*scaleX, fy*scaleY, cx*scaleX, cy*scaleY, pose*deviceTColorCamera_, 0, cv::Size(width, height));
|
||||
}
|
||||
ArImage_release(depthImage);
|
||||
}
|
||||
}
|
||||
|
||||
ArCamera_release(ar_camera);
|
||||
|
||||
@@ -60,13 +60,16 @@ public:
|
||||
std::vector<cv::Point3f> * kpts3D = 0);
|
||||
|
||||
public:
|
||||
CameraARCore(void* env, void* context, void* activity, bool smoothing = false);
|
||||
CameraARCore(void* env, void* context, void* activity, bool depthFromMotion = false, bool smoothing = false);
|
||||
virtual ~CameraARCore();
|
||||
|
||||
bool uvsInitialized() const {return uvs_initialized_;}
|
||||
const float* uvsTransformed() const {return transformed_uvs_;}
|
||||
void getVPMatrices(glm::mat4 & view, glm::mat4 & projection) const {view=viewMatrix_; projection=projectionMatrix_;}
|
||||
|
||||
void updateOcclusionImage(bool enabled) {updateOcclusionImage_ = enabled;}
|
||||
const cv::Mat & getOcclusionImage(CameraModel * model=0) const {if(model)*model=occlusionModel_; return occlusionImage_; }
|
||||
|
||||
virtual void setScreenRotationAndSize(ScreenRotation colorCameraToDisplayRotation, int width, int height);
|
||||
|
||||
virtual bool init(const std::string & calibrationFolder = ".", const std::string & cameraName = "");
|
||||
@@ -101,6 +104,11 @@ private:
|
||||
bool uvs_initialized_ = false;
|
||||
glm::mat4 viewMatrix_;
|
||||
glm::mat4 projectionMatrix_;
|
||||
|
||||
bool updateOcclusionImage_;
|
||||
cv::Mat occlusionImage_;
|
||||
CameraModel occlusionModel_;
|
||||
bool depthFromMotion_;
|
||||
};
|
||||
|
||||
} /* namespace rtabmap */
|
||||
|
||||
@@ -690,7 +690,7 @@ bool RTABMapApp::startCamera(JNIEnv* env, jobject iBinder, jobject context, jobj
|
||||
else if(cameraDriver_ == 1)
|
||||
{
|
||||
#ifdef RTABMAP_ARCORE
|
||||
camera_ = new rtabmap::CameraARCore(env, context, activity, smoothing_);
|
||||
camera_ = new rtabmap::CameraARCore(env, context, activity, depthFromMotion_, smoothing_);
|
||||
#else
|
||||
UERROR("RTAB-Map is not built with ARCore support!");
|
||||
#endif
|
||||
@@ -1113,15 +1113,24 @@ int RTABMapApp::Render()
|
||||
const float* uvsTransformed = 0;
|
||||
glm::mat4 arProjectionMatrix(0);
|
||||
glm::mat4 arViewMatrix(0);
|
||||
rtabmap::Mesh occlusionMesh;
|
||||
if((cameraDriver_ == 1 || cameraDriver_ == 2) && camera_!=0)
|
||||
{
|
||||
boost::mutex::scoped_lock lock(cameraMutex_);
|
||||
if(camera_!=0)
|
||||
{
|
||||
camera_->spinOnce();
|
||||
#ifdef RTABMAP_ARCORE
|
||||
if(cameraDriver_ == 1)
|
||||
{
|
||||
((rtabmap::CameraARCore*)camera_)->updateOcclusionImage(!visualizingMesh_ && main_scene_.GetCameraType() == tango_gl::GestureCamera::kFirstPerson);
|
||||
}
|
||||
#endif
|
||||
|
||||
camera_->spinOnce();
|
||||
|
||||
#ifdef RTABMAP_ARCORE
|
||||
if(cameraDriver_ == 1)
|
||||
{
|
||||
if(main_scene_.background_renderer_ == 0)
|
||||
{
|
||||
main_scene_.background_renderer_ = new BackgroundRenderer();
|
||||
@@ -1131,10 +1140,29 @@ int RTABMapApp::Render()
|
||||
{
|
||||
uvsTransformed = ((rtabmap::CameraARCore*)camera_)->uvsTransformed();
|
||||
((rtabmap::CameraARCore*)camera_)->getVPMatrices(arViewMatrix, arProjectionMatrix);
|
||||
//main_scene_.background_renderer_->Draw(uvsTransformed);
|
||||
}
|
||||
#endif
|
||||
if(!visualizingMesh_ && main_scene_.GetCameraType() == tango_gl::GestureCamera::kFirstPerson)
|
||||
{
|
||||
rtabmap::CameraModel occlusionModel;
|
||||
cv::Mat occlusionImage = ((rtabmap::CameraARCore*)camera_)->getOcclusionImage(&occlusionModel);
|
||||
|
||||
if(occlusionModel.isValidForProjection())
|
||||
{
|
||||
pcl::IndicesPtr indices(new std::vector<int>);
|
||||
pcl::PointCloud<pcl::PointXYZ>::Ptr cloud = rtabmap::util3d::cloudFromDepth(occlusionImage, occlusionModel, 1, 0, 0, indices.get());
|
||||
cloud = rtabmap::util3d::transformPointCloud(cloud, rtabmap::opengl_world_T_rtabmap_world*occlusionModel.localTransform());
|
||||
occlusionMesh.cloud.reset(new pcl::PointCloud<pcl::PointXYZRGB>());
|
||||
pcl::copyPointCloud(*cloud, *occlusionMesh.cloud);
|
||||
occlusionMesh.indices = indices;
|
||||
occlusionMesh.polygons = rtabmap::util3d::organizedFastMesh(cloud, 1.0*M_PI/180.0, false, meshTrianglePix_);
|
||||
}
|
||||
else
|
||||
{
|
||||
UERROR("invalid occlusionModel: %f %f %f %f %dx%d", occlusionModel.fx(), occlusionModel.fy(), occlusionModel.cx(), occlusionModel.cy(), occlusionModel.imageWidth(), occlusionModel.imageHeight());
|
||||
}
|
||||
}
|
||||
}
|
||||
#endif
|
||||
}
|
||||
}
|
||||
|
||||
@@ -1820,7 +1848,7 @@ int RTABMapApp::Render()
|
||||
|
||||
fpsTime.restart();
|
||||
main_scene_.setFrustumVisible(camera_!=0);
|
||||
lastDrawnCloudsCount_ = main_scene_.Render(uvsTransformed, arViewMatrix, arProjectionMatrix);
|
||||
lastDrawnCloudsCount_ = main_scene_.Render(uvsTransformed, arViewMatrix, arProjectionMatrix, occlusionMesh);
|
||||
if(renderingTime_ < fpsTime.elapsed())
|
||||
{
|
||||
renderingTime_ = fpsTime.elapsed();
|
||||
@@ -2106,6 +2134,14 @@ void RTABMapApp::setSmoothing(bool enabled)
|
||||
}
|
||||
}
|
||||
|
||||
void RTABMapApp::setDepthFromMotion(bool enabled)
|
||||
{
|
||||
if(depthFromMotion_ != enabled)
|
||||
{
|
||||
depthFromMotion_ = enabled;
|
||||
}
|
||||
}
|
||||
|
||||
void RTABMapApp::setAppendMode(bool enabled)
|
||||
{
|
||||
if(appendMode_ != enabled)
|
||||
|
||||
@@ -110,6 +110,7 @@ class RTABMapApp : public UEventsHandler {
|
||||
void setCameraColor(bool enabled);
|
||||
void setFullResolution(bool enabled);
|
||||
void setSmoothing(bool enabled);
|
||||
void setDepthFromMotion(bool enabled);
|
||||
void setAppendMode(bool enabled);
|
||||
void setDataRecorderMode(bool enabled);
|
||||
void setMaxCloudDepth(float value);
|
||||
@@ -183,6 +184,7 @@ class RTABMapApp : public UEventsHandler {
|
||||
bool trajectoryMode_;
|
||||
bool rawScanSaved_;
|
||||
bool smoothing_;
|
||||
bool depthFromMotion_;
|
||||
bool cameraColor_;
|
||||
bool fullResolution_;
|
||||
bool appendMode_;
|
||||
|
||||
@@ -512,6 +512,19 @@ Java_com_introlab_rtabmap_RTABMapLib_setSmoothing(
|
||||
}
|
||||
}
|
||||
JNIEXPORT void JNICALL
|
||||
Java_com_introlab_rtabmap_RTABMapLib_setDepthFromMotion(
|
||||
JNIEnv*, jclass, jlong native_application, bool enabled)
|
||||
{
|
||||
if(native_application)
|
||||
{
|
||||
return native(native_application)->setDepthFromMotion(enabled);
|
||||
}
|
||||
else
|
||||
{
|
||||
UERROR("native_application is null!");
|
||||
}
|
||||
}
|
||||
JNIEXPORT void JNICALL
|
||||
Java_com_introlab_rtabmap_RTABMapLib_setCameraColor(
|
||||
JNIEnv*, jclass, jlong native_application, bool enabled)
|
||||
{
|
||||
|
||||
@@ -367,7 +367,7 @@ bool intersectFrustumAABB(
|
||||
}
|
||||
|
||||
//Should only be called in OpenGL thread!
|
||||
int Scene::Render(const float * uvsTransformed, glm::mat4 arViewMatrix, glm::mat4 arProjectionMatrix) {
|
||||
int Scene::Render(const float * uvsTransformed, glm::mat4 arViewMatrix, glm::mat4 arProjectionMatrix, const rtabmap::Mesh & occlusionMesh) {
|
||||
UASSERT(gesture_camera_ != 0);
|
||||
|
||||
if(currentPose_ == 0)
|
||||
@@ -458,7 +458,7 @@ int Scene::Render(const float * uvsTransformed, glm::mat4 arViewMatrix, glm::mat
|
||||
|
||||
UTimer timer;
|
||||
|
||||
bool onlineBlending = blending_ && gesture_camera_->GetCameraType()!=tango_gl::GestureCamera::kTopOrtho && mapRendering_ && meshRendering_ && cloudsToDraw.size()>1;
|
||||
bool onlineBlending = (renderBackgroundCamera && occlusionMesh.cloud.get() && occlusionMesh.cloud->size()) || (blending_ && gesture_camera_->GetCameraType()!=tango_gl::GestureCamera::kTopOrtho && mapRendering_ && meshRendering_ && cloudsToDraw.size()>1);
|
||||
if(onlineBlending && fboId_)
|
||||
{
|
||||
// set the rendering destination to FBO
|
||||
@@ -468,11 +468,19 @@ int Scene::Render(const float * uvsTransformed, glm::mat4 arViewMatrix, glm::mat
|
||||
glClearColor(1, 1, 1, 1);
|
||||
glClear(GL_DEPTH_BUFFER_BIT | GL_COLOR_BUFFER_BIT);
|
||||
|
||||
// Draw scene
|
||||
for(std::vector<PointCloudDrawable*>::const_iterator iter=cloudsToDraw.begin(); iter!=cloudsToDraw.end(); ++iter)
|
||||
if(renderBackgroundCamera)
|
||||
{
|
||||
// set large distance to cam to use low res polygons for fast processing
|
||||
(*iter)->Render(projectionMatrix, viewMatrix, meshRendering_, pointSize_, false, false, 999.0f);
|
||||
PointCloudDrawable drawable(occlusionMesh);
|
||||
drawable.Render(projectionMatrix, viewMatrix, true, pointSize_, false, false, 999.0f);
|
||||
}
|
||||
else
|
||||
{
|
||||
// Draw scene
|
||||
for(std::vector<PointCloudDrawable*>::const_iterator iter=cloudsToDraw.begin(); iter!=cloudsToDraw.end(); ++iter)
|
||||
{
|
||||
// set large distance to cam to use low res polygons for fast processing
|
||||
(*iter)->Render(projectionMatrix, viewMatrix, meshRendering_, pointSize_, false, false, 999.0f);
|
||||
}
|
||||
}
|
||||
|
||||
// back to normal window-system-provided framebuffer
|
||||
@@ -512,6 +520,10 @@ int Scene::Render(const float * uvsTransformed, glm::mat4 arViewMatrix, glm::mat
|
||||
if(renderBackgroundCamera)
|
||||
{
|
||||
background_renderer_->Draw(uvsTransformed);
|
||||
|
||||
//To debug occlusion image:
|
||||
//PointCloudDrawable drawable(occlusionMesh);
|
||||
//drawable.Render(projectionMatrix, viewMatrix, true, pointSize_, false, false, 999.0f);
|
||||
}
|
||||
|
||||
if(!currentPose_->isNull())
|
||||
|
||||
@@ -72,13 +72,14 @@ class Scene {
|
||||
// frame's timestamp.
|
||||
// @param: point_cloud_vertices, point cloud's vertices of the current point
|
||||
// frame.
|
||||
int Render(const float * uvsTransformed = 0, glm::mat4 arViewMatrix = glm::mat4(0), glm::mat4 arProjectionMatrix=glm::mat4(0));
|
||||
int Render(const float * uvsTransformed = 0, glm::mat4 arViewMatrix = glm::mat4(0), glm::mat4 arProjectionMatrix=glm::mat4(0), const rtabmap::Mesh & occlusionMesh=rtabmap::Mesh());
|
||||
|
||||
// Set render camera's viewing angle, first person, third person or top down.
|
||||
//
|
||||
// @param: camera_type, camera type includes first person, third person and
|
||||
// top down
|
||||
void SetCameraType(tango_gl::GestureCamera::CameraType camera_type);
|
||||
tango_gl::GestureCamera::CameraType GetCameraType() const {return gesture_camera_->GetCameraType();}
|
||||
|
||||
void SetCameraPose(const rtabmap::Transform & pose); // opengl camera
|
||||
rtabmap::Transform GetCameraPose() const {return currentPose_!=0?*currentPose_:rtabmap::Transform();}
|
||||
|
||||
Reference in New Issue
Block a user