klischa/astra2

Purpose and entry points

AstraScanner2 is a Windows C++17 desktop application for RGB-D 3D scanning, point-cloud processing, mesh reconstruction, multi-scan registration, and file export. The executable is assembled by CMake at CMakeLists.txt; application startup is src/main.cpp, which creates the Qt UI shell in src/gui/MainWindow.cpp.

UI orchestration layer

Qt6 UI is intentionally thin: binders assemble tab/menu/status widgets while controllers own user workflows, state transitions, and asynchronous orchestration. src/gui/MainWindow.cpp connects the shell to src/controllers/ProjectUiController.cpp, src/controllers/CaptureSessionController.cpp, and src/controllers/ProcessingController.cpp. src/controllers/ViewerCoordinator.cpp coordinates main and processing visualizations and switches cloud/mesh displays.

Capture and scan-session pipeline

A scan session starts through CaptureSessionController, which manages a worker/thread boundary around src/capture/CaptureWorker.cpp. The worker acquires RGB/depth frames through src/capture/AstraCamera.cpp, using OpenNI2 for an Orbbec Astra Pro when available or synthetic emulation otherwise; depth is converted to point-cloud data via src/capture/DepthCloudConverter.h. src/controllers/ScanWorkflow.cpp and src/controllers/TurntableController.cpp coordinate manual and timed turntable scan flows.

Pose and coverage tracking

Captured frames feed one of two pose-estimation paths. Marker tracking is defined behind src/marker_tracker/IMarkerDetector.h; src/marker_tracker/MarkerTracker.cpp selects CCTag/circular detection and uses src/marker_tracker/MarkerMap.cpp to derive object pose from known markers. Neural tracking uses the bundled ONNX model at models/pose_regressor.onnx, executed by src/tracking/NeuralTracker.cpp and coordinated/validated in src/tracking/NeuralTrackingPipeline.cpp. src/tracking/SurfelCoverageTracker.cpp evaluates scan coverage/quality. Neural inference can use ONNX Runtime CUDA execution, with guard logic in src/tracking/NeuralTrackingGuards.cpp.

Processing and reconstruction services

ProcessingController delegates computational work to services rather than UI code. src/services/CloudFiltersService.cpp performs cloud cleanup; the compatibility/filter facade is src/filters/PointCloudFilters.cpp. CUDA-specific implementations and wrappers are isolated in src/filters/PointCloudFiltersCUDA.cu and src/filters/PointCloudFiltersCUDA_wrapper.cpp, with CPU PCL fallback. src/services/RegistrationService.cpp performs ICP-based scan alignment/merging. src/services/ReconstructionService.cpp produces Poisson meshes, and src/services/MeshProcessingService.cpp handles smoothing, decimation, and hole filling.

Persistence and output boundary

src/project/ProjectManager.cpp owns project lifecycle and durable metadata/cloud state, separating workspace persistence from processing. src/export/ExportManager.cpp is the outward format boundary for PLY, PCD, STL, and OBJ. Application-wide operational settings, including capture/filter/tracking parameters, are centralized in src/settings/SettingsManager.cpp; marker-disk configuration is supplied at data/marker_disk_config.json.

Runtime and infrastructure

The native runtime combines Qt6, PCL, VTK/OpenGL visualization, OpenCV, CCTag, Eigen, and ONNX Runtime, declared and linked via CMakeLists.txt and vcpkg.json. CUDA is optional and targets NVIDIA GPUs; OpenNI2 is optional hardware infrastructure for Astra Pro. Deployment must include Qt plugins, vcpkg/OpenNI2/ONNX DLLs, model assets, and data/, as configured by the build system.