Skip to content

bbox_object_locator#

Inner Workings / Algorithms#

This node produces 3D object detection from a 2D image bounding box (bbox) through the following steps:

  1. Back-project the bottom-center pixel of a bbox onto the ground plane (z=0 in the target coordinate system). This point is used as the object's bottom center.
  2. Back-project the top-center pixel of the bbox to a point near the one computed in step 1 for height estimation.
  3. Back-project the bottom-left and bottom-right pixels of the bbox to estimate the object’s width (diameter).
  4. Use these four points to determine the object’s position and size.

Parameters#

Name Type Description Default Range
target_frame string Transform target frame. base_link N/A
detection_target_class.UNKNOWN boolean If true, objects with the 'UNKNOWN' label will be ignored during processing. false N/A
detection_target_class.CAR boolean If true, objects with the 'CAR' label will be ignored during processing. false N/A
detection_target_class.TRUCK boolean If true, objects with the 'TRUCK' label will be ignored during processing. false N/A
detection_target_class.BUS boolean If true, objects with the 'BUS' label will be ignored during processing. false N/A
detection_target_class.TRAILER boolean If true, objects with the 'TRAILER' label will be ignored during processing. false N/A
detection_target_class.MOTORCYCLE boolean If true, objects with the 'MOTORCYCLE' label will be ignored during processing. false N/A
detection_target_class.BICYCLE boolean If true, objects with the 'BICYCLE' label will be ignored during processing. false N/A
detection_target_class.PEDESTRIAN boolean If true, objects with the 'PEDESTRIAN' label will be ignored during processing. true N/A
detection_target_class.ANIMAL boolean If true, objects with the 'ANIMAL' label will be ignored during processing. false N/A
detection_target_class.HAZARD boolean If true, objects with the 'HAZARD' label will be ignored during processing. false N/A
roi_confidence_threshold float Threshold value for whether to trust and use the ROI. If the confidence of an ROI is below this value, it will not be used. 0.8 ≥0.0
≤1.0
roi_truncation_validation.enabled array Flags to enable validation for each ROI. Must have the same number as the input ROIs. [false, false, false, false, false, false, false, false, false, false] N/A
roi_truncation_validation.remove_truncated array If this is true, the object will be removed when its ROI is evaluated as truncated. Must have the same number as the input ROIs. [false, false, false, false, false, false, false, false, false, false] N/A
roi_truncation_validation.image_border_truncation_horizontal_margin_ratio float Margin ratio used to detect whether the ROI is truncated based on the image width. 0.0 ≥0.0
≤0.5
roi_truncation_validation.image_border_truncation_vertical_margin_ratio float Margin ratio used to detect whether the ROI is truncated based on the image height. 0.0 ≥0.0
≤0.5
detection_max_range float Maximum detection range. Objects beyond this distance will not be included in the output [m]. 30.0 ≥0.0
pseudo_height float If height was not able to compute, this value will used as pseudo height of a object [m]. 1.5 ≥0.1
pedestrian_detection_config.width_min float If the estimated width is smaller than this value, it will be clamped to this value [m]. 0.5 ≥0.3
pedestrian_detection_config.width_max float If the estimated width is larger than this value, it will be clamped to this value [m]. 0.8 ≥0.3

Input#

Name Type Description
input/rois<id> tier4_perception_msgs::msg::DetectedObjectsWithFeature 's input ROI
input/camera_info<id> sensor_msgs::msg::CameraInfo 's camera info

Output#

Name Type Description
output/rois<id>/objects autoware_perception_msgs::msg::DetectedObjects The object generated from 's 2D ROI