Detection2DStepSettings< OBJ_DET_2D_TYPE::FEATURE_MATCHING >

Step of the 2D detection pipeline.

Define a set of CorrectionFeature that define features of an object. If all features are uniquely detected in a current oriented bounding box it is optimized such that the distance to these defined features is minimized. Internally an iterative Gauss-Newton algorithm is applied.

Member Functions

virtual OBJ_DET_2D_TYPE type() const override: JSON Type: string +Requirement: Must be OBJ_DET_2D_TYPE::FEATURE_MATCHING;. Type of the 2D detection step.

Member Variables

optimizationParameters: std::vector< FEATURE_MATCHING_OPT_PARAM > JSON Type: vector of string +Requirement: Must not be empty. Each element must be a FEATURE_MATCHING_OPT_PARAM Indicate which parameters are optimized during feature matching.

iterations: int, default: {3} JSON Type: int +Requirement: Must be greater than zero. Number of Gauss-Newton iterations.

features: std::vector< CorrectionFeature > JSON Type: vector of CorrectionFeature +Requirement: Must not be empty. Description of features and their location within each oriented bounding box.

Enumerations

Table 1. FEATURE_MATCHING_OPT_PARAM
Value	Description
`SCALEX`	Scale the oriented bounding box in x direction.
`SCALEY`	Scale the oriented bounding box in y direction.
`TX`	Translate the oriented bounding box in x direction.
`TY`	Translate the oriented bounding box in y direction.
`ANGLE`	Rotate the oriented bounding box.

Table 2. OBJ_DET_2D_TYPE
Value	Description
`SET_FULL`	All current oriented bounding boxes are set to the full image size (top is up). >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SET_FULL>.
`SET_ROI`	All current oriented bounding boxes are set to a specified ROI (top is up). >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SET_ROI>.
`SET_ROI_PIPELINE`	All current oriented bounding boxes are set to the ROI of the pipeline (top is up). >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SET_ROI_PIPELINE>.
`SET_OTHER`	All current oriented bounding boxes are set to the result of another 2D detection pipeline. >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SET_OTHER>.
`SET_NETWORK`	Applies an object detection network to all current oriented bounding boxes. Returns all detected >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SET_NETWORK>.
`FILTER_CLASSIFICATION`	Applies a classification filter to all current oriented bounding boxes. Filters out all boxes >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::FILTER_CLASSIFICATION>.
`TRANSLATE`	Translate all current oriented bounding boxes relative to their current orientation and size. >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::TRANSLATE>.
`ROTATE_ANGLE`	Rotate all current oriented bounding boxes by a fixed angle around a fixed point that is relative >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::ROTATE_ANGLE>.
`SCALE`	Scale all current oriented bounding boxes by a fixed value from a fixed point that is relative >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SCALE>.
`SCALE_TO`	Scale all current oriented bounding boxes to a fixed aspect ratio from a fixed point that is relative >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SCALE_TO>.
`SCALE_ASPECT`	Scale all current oriented bounding boxes to a fixed aspect ratio from a fixed point that is relative >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::SCALE_ASPECT>.
`REORIENT_ASPECT`	Modify all oriented bounding boxes "up-direction", such that the long or short size becomes the top. Note that all box corners stay the same. >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::REORIENT_ASPECT>.
`REORIENT_CLASSIFICATION`	Modify all oriented bounding boxes "up-direction" using a classification network. The network classes must be "Links", "Rechts", "Oben", "Unten". >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::REORIENT_CLASSIFICATION>.
`FEATURE_MATCHING`	Define a set of CorrectionFeature that define features of an object. If all features are uniquely detected in a current oriented bounding box it is optimized such that the distance to these defined features is minimized. Internally an iterative Gauss-Newton algorithm is applied. >>> See Detection2DStepSettings<OBJ_DET_2D_TYPE::FEATURE_MATCHING>.
`ADD_INSTANCES_FROM_FILE`	See Detection2DStepSettings<OBJ_DET_2D_TYPE::ADD_INSTANCES_FROM_FILE>.