How hku-mars/m-detector works
M-detector is a ROS LiDAR moving-point detection package. Its principal workflow consumes point clouds and odometry, applies occlusion/depth-map tests to classify points, and publishes point- and frame-oriented results. Separate tools cluster dynamic points, visualize saved predictions, and calculate IoU/recall against dataset labels. The README documents the launch-level workflow; source excerpts show filtering and publishing details, but do not establish all connections among the filter, clustering, and evaluation executables.
Main entry points: DynObjFilter.cpp Occlusion filter · DynObjCluster.cpp Dynamic clustering · cal_recall_multi.cpp Recall and IoU · display_prediction.cpp Prediction viewer
Components 16
Online detection
- Occlusion filter
src/DynObjFilter.cpp - Depth-map state
include/m-detector/DynObjFilter.h - Detection topics
src/DynObjFilter.cpp - Sensor configurations
config
Object clustering
- Dynamic clustering
src/DynObjCluster.cpp - Cluster interface
include/m-detector/DynObjCluster.h
Results tools
- Saved predictions
- Dataset labels
- Recall and IoU
src/cal_recall_multi.cpp - Prediction viewer
src/display_prediction.cpp - ROS visualization
src/display_prediction.cpp
Shared algorithms
- Voxel clustering
include/cluster_predict/voxel_cluster.h - DBSCAN clustering
include/cluster_predict/dbscan.h - Box geometry and IoU
include/cluster_predict/bbox.h - Disk neighborhood
include/cluster_predict/EA_disk.h
Other components
- LiDAR and odometry
Connections 13
- LiDAR and odometry → Occlusion filter: supplies streams
README.md - Occlusion filter → Depth-map state: updates maps
src/DynObjFilter.cpp - Occlusion filter → Detection topics: publishes results
src/DynObjFilter.cpp - Sensor configurations → Occlusion filter: configures detection
- Dynamic clustering → Voxel clustering: uses voxel map
src/DynObjCluster.cpp - Dynamic clustering → Disk neighborhood: includes helper
src/DynObjCluster.cpp - Cluster interface → Dynamic clustering: declares interface
- Saved predictions → Recall and IoU: provides predictions
src/cal_recall_multi.cpp - Dataset labels → Recall and IoU: provides ground truth
src/cal_recall_multi.cpp - Saved predictions → Prediction viewer: provides predictions
src/display_prediction.cpp - LiDAR and odometry → Prediction viewer: provides point cloud
src/display_prediction.cpp - Prediction viewer → ROS visualization: publishes views
src/display_prediction.cpp - Recall and IoU → Saved predictions: writes metrics