ChoiWooCheol / QT-LiDAR-Object-Detection

QT (Quad-Tree Segmentation)

QT-LiDAR-Object-Detection

• QT (Quad-Tree Segmentation).
• Upgrade version of LiDAR-obstacle-dectection Repository.
• Realtime object recognization, using only LiDAR.
• Available for real-time self-driving systems.
• More powerful than euclidean clustering detection

Hardware

• HYUNDAI i30
• Ouster OS1 64 channel LiDAR
• Intel Core i5-8250U, 3.4Ghz
• 16G RAM
• Geforce 1050GTX

What is changed content

• Add include files.
• Existing method did not provide minium size bounding box, but this version is providing.
• Use corvarience of points, and calculate Quaternion and Rotation information of bounding box.
• bounding box's pose has orientation values.
• if you use vector map, can change cluster size (do not generte bounding box of static objects in vectormap)

Input topic

• '/points_raw'

Output topic

• '/detected_boxes'
• '/obb_cluster'
• '/obb_boxes'

MODE

• OBB MODE : Bounding boxes are minimum size and have orientation values
• AABB MODE : Bounding boxes are not minimun size and do not have orientation. but a little bit fast
• USE_VECTORMAP MODE : be going to add

Run

$ roslaunch lidar_detect qt_detect_launch.launch

Result

1. Existing method

• no rotation
• no minimum size box

2. QT-detect output

• has orientation
• minimum size box for clustering obj