diff --git a/params/iri_ground_segmentation_no_random_forest.yaml b/params/iri_ground_segmentation_no_random_forest.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3d8b5de2c04e11fe6f722bc5f26315968a34963f --- /dev/null +++ b/params/iri_ground_segmentation_no_random_forest.yaml @@ -0,0 +1,51 @@ +iri_ground_segmentation: { + rate: 10.00, + + # Robot related geometric parameters + sensor_height: 1.73, ## Ground is expected to be at z = -1*sensor_height + robot_height: 2.00, ## All obstacle points above this value (w.r.t the z ground predicted by the algorithm) are treated as overhanging obstacles + + # Parameters affecting the exploration of the pointcloud + ROI_delta_x_and_y: 3.0, ## This value sets the size of the ROIs: ROI_size = (2*ROI_delta_x_and_y)^2 + ROI_shadow_area: 5.5, ## This value is the same that the above, but only used in the root vertex to overcome the shadow area + + ground_reference_search_resolution_deg: 40.0, #12.00, ## It is the angular resolution when generating new ground references (graph nodes), + ## it will affect the number of nodes in the graph: lower values generate more nodes + + elevation_grid_resolution: 2.1, #0.333, ## This value is used to create the "elevation point cloud" which is a reduction of the original pointcloud, where + ## only the lowest z values in each cell are stored (along with a vector of indexes pointing to the remaining points + ## in the cell, so that the original pointcloud can be recovered or labeled using the info in the "elevation cloud"). + ## Big values can speed up the algorithm but generates a lower resolution ground model, on the other hand, small values + ## can produce outliers because we would like to have only ground points in the elevation cloud (however these outliers + ## are usually succesfully rejected by the mahalanobis threshold and the "prior" information) + + # Kalman filter noise parameters + ## initial uncertainties + z_initial_std_dev: 0.05, ## Uncertainty in z = -1*sensor_height + initial_angular_std_dev_deg: 1.5, ## Used to initialize x_slope and y_slope variances + + ## propagation additive noises + propagation_z_additive_noise_per_meter: 0.01, ## Uncertainty added to var_z as a function of the distance with the parent vertex + propagation_additive_noise_deg_per_meter: 0.4, ## Uncertainty added to x_slope and y_slope as a function of the distance with the parent vertex + ## (it is expressed in degrees per meter, but inside the code it is converted to slope per meter) + # measurement noise + z_observation_std_dev: 0.15, ## Uncertainty in the Lidar observations + + # threshold for outlier rejection and classification + mahalanobis_threshold: 2.7, ## To classify points as obstacles or ground (a small value will cause false obstacle points, + ## and a big value will increase the number of false ground points) + + # labeling parameters + number_of_references_used_for_labelling: 0, ## used to evaluate each elevation point cloud from different POVs // NEW!! value of zero enables fast labelling mode!! \m/ + max_pred_std_dev_for_labelling: 0.5, ## ONLY IN USE TO GIVE COLOUR TO DENSE RECONSTRUCTION + score_threshold: 0.0, ## for assigning ground class label: one means maha. dist. equal to zero, zero means mahalanobis dist equal to maha. thres + classify_not_labeled_points_as_obstacles: true, ## when a point has no reference satisfying the max_pred_std_dev_for_labelling threshold we can leave as unknown or label it as obstacle + ground_threshold_in_not_analyzed_areas: 0.1, ## when it is not possible to make a local analysis of the data, we will use the lowest point (the one in elevation_cloud) as ground height, and + ## all the points above it and below this threshold will be classified as ground + discard_not_connected_references_for_labelling: false, ## NOT IN USE + + # visualization and debug parameters + measure_performance: false, ## (feature still not debugged) Flag to measure number of execution and execution times of the different functions of the algorithm + show_dense_reconstruction: false, ## To show a dense ground surface reconstruction using the predictions of the ground mode (colored using the std_dev of z coordinate) + ## or alternatively the "elevation point cloud" (useful for parameter tunning) +}