perception-lidar-detection

Introduction

Perform 3D object detection based on the point cloud, and output the position, size, and orientation of the detected objects. Apollo provides 4 lidar detection models: centerpoint, maskpillars, pointpillars, cnnseg.

Structure

├── lidar_detection
    ├── conf            // component config file
    ├── dag             // component dag file
    ├── data            // model config files
    ├── detector        // lidar model
    │   ├── center_point_detection   // CenterPoint model code
    │   ├── cnn_segmentation         // CNNSeg model code
    │   ├── mask_pillars_detection   // MaskPillars model code
    │   └── point_pillars_detection  // PointPillars model code
    ├── interface       // definition of BaseLidarDetector
    ├── proto           // definition of data structure
    ├── lidar_detection_component.cc // component inference
    ├── lidar_detection_component.h
    ├── cyberfile.xml   // package configs
    ├── README.md
    └── BUILD

Modules

LidarDetectionComponent

apollo::perception::lidar::LidarDetectionComponent

Input

Note: The input channel is structure type data. The default trigger channel is /perception/lidar/pointcloud_ground_detection. The detailed input channel information is in modules/perception/lidar_detection/dag/lidar_detection.dag file. By default, the upstream components of the messages received by the component include pointcloud_ground_detection.

Output

Note: The output channel is structure type data. The message is defined in the modules/perception/common/onboard/inner_component_messages/lidar_inner_component_messages.h file. The output channel message data can be subscribed by components in the same process. The detailed output channel information is in modules/perception/lidar_detection/conf/lidar_detection_config.pb.txt file.

How to use

1. Add vehicle params configs to modules/perception/data/params，keep frame_id and sensor_name consistent, start the tf

cyber_launch start modules/transform/launch/static_transform.launch

2. Select the lidar model，download the model files and put them into modules/perception/data/models using amodel tool.

amodel install https://xxx.zip

3. Modify the modules/perception/lidar_detection/dag/lidar_detection.dag

• config_file_path: path of config path
• reader channel: the name of input channel

4. Modify modules/perception/lidar_detection/conf/lidar_detection_config.pb.txt

• output_channel_name: the name of output channel
• sensor_name: sensor name
• use_object_builder: whether use object builder
• plugin_param
  • name： the name of lidar model
  • config_path: the path of model config
  • config_file: the name of model config

5. start the lidar component

cyber_launch start modules/perception/launch/perception_lidar.launch

Reference

1. Center-based 3D Object Detection and Tracking
2. PointPillars: Fast Encoders for Object Detection from Point Clouds