Library Integration for DIS - Preparing the Runtime Settings
 | This page page is under construction. |
Runtime Settings Preparation
Before creating the DigitalStabilizer, the application must configure both the stabilization parameters and the runtime settings required by the selected backend.
The stabilization parameters define the behavior of the pipeline, such as optical flow configuration, smoothing settings, and the algorithms used for motion estimation and image transformation.
Common Stabilization Parameters
The example first creates and configures the optical flow parameters:
auto optical_flow_params =
std::make_shared<rvs::OpticalFlowParams>();
optical_flow_params->feature_redetect_interval =
opts.harris_refresh;
optical_flow_params->feature_max_corners =
opts.max_corners;
optical_flow_params->opt_flow_num_levels =
opts.pyramid_levels;
These parameters are then attached to the DigitalStabilizer configuration:
auto digital_params =
std::make_shared<
rvs::DigitalStabilizerParams>();
digital_params->smoothing_frames =
opts.smoothing_frames;
digital_params->crop_margin =
opts.crop_margin;
digital_params->optical_flow_params =
optical_flow_params;
digital_params->smoothing_algorithm =
rvs::SmoothingAlgorithms::kGaussian;
The resulting configuration hierarchy is:
DigitalStabilizerParams
├── OpticalFlowParams
├── Smoothing Configuration
└── Transform Configuration
HostAllocator (OpenCV Backend)
For the OpenCV backend, no special runtime configuration is required. The example selects the OpenCV implementations for both optical flow and geometric transformations:
digital_params->optical_flow_algorithm =
rvs::OpticalFlowAlgorithms::
kPyrLKOpenCV;
digital_params->transform_algorithm =
rvs::TransformAlgorithms::
kTransformOpenCV;
No runtime settings object is created:
std::shared_ptr<
rvs::IRuntimeSettings>
runtime_settings = nullptr;
The stabilizer is then created directly:
rvs::DigitalStabilizer stabilizer(
digital_params,
runtime_settings);
CudaAllocator (CUDA Backend)
For the CUDA backend, the example selects the CUDA implementations:
digital_params->optical_flow_algorithm =
rvs::OpticalFlowAlgorithms::
kPyrLKCUDA;
digital_params->transform_algorithm =
rvs::TransformAlgorithms::
kTransformCUDA;
Unlike the OpenCV backend, CUDA requires additional runtime configuration. The example creates a CUDA stream and stores it in a CUDA-specific runtime settings object:
auto cuda_runtime_settings =
std::make_shared<
rvs::CudaRuntimeSettings>();
cudaStreamCreateWithFlags(
&cuda_stream,
cudaStreamNonBlocking);
cuda_runtime_settings->cuda_stream =
cuda_stream;
cuda_runtime_settings->async = true;
The CUDA runtime settings are then exposed through the common runtime settings interface:
std::shared_ptr<
rvs::IRuntimeSettings>
runtime_settings =
cuda_runtime_settings;
The resulting configuration hierarchy is:
IRuntimeSettings
↑
CudaRuntimeSettings
├── CUDA Stream
└── Async Execution Mode
Finally, the stabilizer is created using the configured parameters and runtime settings:
rvs::DigitalStabilizer stabilizer(
digital_params,
runtime_settings);
