Pose Detector#
The last stage of the dynamic use cases Vision component of RVC is the Pose Detector.
Its role is to acquire the RVC Vision Messages RotateBBList message subscribing from the object detection component and the PointCloud2 message from Intel® RealSense™ camera node
With these information, using the pointcloud representation of the objects from storage, the Pose Detector will try to find instances of the idea object in the camera pointcloud scene.
The corresponding mesh in pcd format is used according to the RotatedBB class id, so if the Object Detection found a cube, the pose detector will use the cube.pcd to try to find the cube in the Intel® RealSense™ camera with the Point Cloud Library (PCL) scene.
The package providing the object pcd files is specified in the yaml file as rvc_use_case_binaries
as described below
This node accept following parameters:
class_name_array: filenames to load from file, pose detector will append .pcd and search in the rvc_use_case_binaries package, subdirectory pcd_objects
rvc_use_case_binaries: package name providing the needed binaries as per previous parameters
The following parameters fine tune the internal pointcloud ransac algorithm aligning the mesh loaded from storage to the Intel® RealSense™ camera’s point cloud output. For further information, refer to SampleConsensusPrerejective. For a sample explanation, refers to PCL Alignment Prerejective
downsampling
downsample_clouds
normal_search_radius
fpfh_search_radius
stddev_mul_threshold
z_threshold
inlier_fraction
similarity_threshold
max_corresp_randomness
max_corresp_distance
max_iterations
num_samples
icp_max_corresp_distance
icp_xform_epsilon
icp_fitness_epsilon
icp_max_iterations
Default Config file#
class_name_array: [ 'bolt', 'gear', 'nut', 'cube' ]
downsampling: 0.0055
downsample_clouds: 0.0055
normal_search_radius: 0.005
fpfh_search_radius: 0.05
stddev_mul_threshold: 0.0
z_threshold: 0.66
inlier_fraction: 0.47
similarity_threshold: 0.86
max_corresp_randomness: 214.0
max_corresp_distance: 0.1
max_iterations: 1000
num_samples: 3
icp_max_corresp_distance: 0.1
icp_xform_epsilon: 0.000000001
icp_fitness_epsilon: 0.000006
icp_max_iterations: 50
rvc_use_case_binaries: rvc_use_case_binaries