How socone-tintran/perception works
Main Components
a. Sensor input modules (although not directly in this repo, the modules process data from sensors such as lidar, radar, camera, and map information).
b. Core perception modules – each package (e.g., autoware_bytetrack, autoware_cluster_merger, autoware_compare_map_segmentation, autoware_ground_segmentation, etc.) represents a building block in the perception pipeline.
c. Data fusion and transformation nodes – packages such as autoware_image_projection_based_fusion, autoware_lidar_transfusion, and autoware_multi_object_tracker.
d. Object validation, tracking, and filtering – many packages offer filtering (e.g., occupancy grid map filters, noise filters), data association, and multi-object tracking.
e. Utility libraries – shared utilities such as in perception_utils and common include folders that coordinate message conversion, data synchronization, and more.
Relationships and Interactions
b. Subsequent branches for segmentation, clustering, object detection, instance segmentation, and fusion.
c. The outputs of these nodes are provided to trackers and object merger modules, with further processing like validation and filtering.
e. Additionally, configuration and launch directories indicate that the system is deployed using ROS launch files, so you may want to include an “orchestration layer” showing how the nodes are launched and interconnect.
Architectural Patterns and Principles
a. Modularity: Each package is self-contained with its own configuration, libraries, schema, and tests. This emphasizes a microservices-like architecture.
c. Separation of concerns: Different packages handle specific tasks (e.g., segmentation vs. tracking vs. fusion vs. noise filtering). This separation is key in the design.
d. Extensibility and configuration-driven paradigms: Many packages include param.yaml, schema files, and launch files. Note these as dynamic configuration and design-time plug points.
Additional Guidance
a. A high-level view (sensor inputs → preprocessing → perception modules → fusion/tracking → output) and then add layers of detail about which package provides which functionality.