Statistics of latest Autoware's Whole Perception Pipeline Latency (camera_lidar_fusion mode)

[cyn-liu]

Checklist

• I've read the contribution guidelines.
• I've searched other issues and no duplicate issues were found.
• I've agreed with the maintainers that I can plan this task.

Description

Previously, we statistics the perception pipeline latency of lidar only mode and analyzed the time consumption of multiple main perception nodes.
#8752

Now, we plan to statistics the perception pipeline latency of camera_lidar_fusion mode and analyze the time consumption of the main perception nodes.

Environment:
Device: DELL precison3581
OS: Ubuntu22.04
GPU: NVIDIA RTX A1000 6GB Laptop
CPU: intel CORE i7

Purpose

In order to get a detailed of each node and the overall time consumption of the perception pipeline, and to facilitate developers in improving the performance of the perception module.

Possible approaches

1. Using debug/pipeline_latency_ms topics to get the time consumption of the perception pipeline.

2. Custom topics to get time consumption of certain perception nodes.

Definition of done

None

[cyn-liu]

Autoware provides debug/processing_time_ms to display processing times for the node.

Perception module publish the following debug/processing_time_ms debug information, we subscribe to these topics and analyzed them to evaluate the time consumption of each node of perception and entire pipeline.

/perception/obstacle_segmentation/crop_box_filter/debug/processing_time_ms 
/perception/occupancy_grid_map/raw_pc_downsample_filter/debug/processing_time_ms
/perception/obstacle_segmentation/scan_ground_filter/debug/processing_time_ms
/perception/occupancy_grid_map/obstacle_pc_downsample_filter/debug/processing_time_ms
/perception/occupancy_grid_map/pointcloud_based_occupancy_grid_map/debug/processing_time_ms
/perception/obstacle_segmentation/occupancy_grid_map_outlier_filter/debug/processing_time_ms 
/perception/object_recognition/detection/voxel_based_compare_map_filter/debug/processing_time_ms 
/perception/object_recognition/detection/clustering/voxel_grid_based_euclidean_cluster/debug/processing_time_ms
/perception/object_recognition/detection/clustering/shape_estimation/debug/processing_time_ms

/perception/object_recognition/detection/tensorrt_yolox/debug/processing_time_ms 
/perception/object_recognition/detection/clustering/roi_cluster/roi_pointcloud_fusion/debug/processing_time_ms
/perception/object_recognition/detection/clustering/camera_lidar_fusion/roi_cluster_fusion/debug/processing_time_ms 
/perception/object_recognition/detection/clustering/camera_lidar_fusion/low_intensity_cluster_filter_node/debug/processing_time_ms 
/perception/object_recognition/detection/clustering/camera_lidar_fusion/shape_estimation/debug/processing_time_ms 

/perception/object_recognition/detection/centerpoint/lidar_centerpoint/debug/processing_time_ms
/perception/object_recognition/detection/roi_detected_object_fusion/debug/processing_time_ms 
/perception/object_recognition/detection/object_association_merger/debug/processing_time_ms
/perception/object_recognition/tracking/multi_object_tracker/debug/processing_time_ms
/perception/object_recognition/detection/detection_by_tracker/detection_by_tracker/debug/processing_time_ms 
/perception/object_recognition/prediction/map_based_prediction/debug/processing_time_ms 

---

Node Relationship Graph of Autoware's camera_lidar_fusion Mode Perception Pipeline

---