human_detector pkg, Is it 3D detection?

[eeoon]

Before proceeding, is there an existing issue or discussion for this?

• I have done a search for similar issues and discussions.

Description

env, Ubuntu 22.04
ROS distro: Humble

hello, @Yadunund . I try this packages for obstadcle detector with camera. I think this packages for static camera on wall like README.md. I used this package, rmf_human_detector for real_robot with realsense_camera(D455, D435). so I make environment wheel_robot with camera_sensor(ROS) in Simulation(Gazebo) and I check the sensor detect a human and publish /rmf_obstacle. I saw Green Box in rmf_map(rviz). but the /rmf_obstacle coordinate is mismatching rmf_map frame coordinate. I think this package used yolov5 for detect a human, but not provided 3D coordinate about detecting human. Right??

Best,
eeoon.

[Yadunund]

I think this package used yolov5 for detect a human, but not provided 3D coordinate about detecting human. Right??

It tries to approximate 3D pose of objects detected from the camera frame based on some pixel-to-meter heuristics. But I think the right approach would be to rectify depth images and get the mean distance of the pixels within the detected bounding box. The rmf_human_detector_oakd does this.

[eeoon]

HI @Yadunund
Thanks for your comment!

I have tried Realsense But I understand rmf_human_detector_oakd package is based on OAK-D camera. right? Um,, I got this problem solved 3D detection Algorithm. So I have Questions.

1. I think this packages is meant to be a lane_blocker. Detecting Obstacle used lidar, camera. and next step is lane_status will be closed. So I want to Lane_Blocker code's Parameter's meant.(Like, cull_time, tf2_lookup_duration etc..)

2. "rmf_obstacle_ros2" package takes a long time to build.. why this situation happen? um.. I test about 20 minitues just "colcon build --packages-select rmf_obstacle_ros2" I don't know how to solve this problem.

Best
eeoon,