ROS package for sensor processing, object detection, tracking, and evaluation using the KITTI Vision Benchmark dataset.
SARosPerceptionKitti is a ROS package that implements a complete perception pipeline for autonomous driving research using the KITTI Vision Benchmark dataset. It processes synchronized sensor data (LiDAR and cameras), detects objects like cars and pedestrians, tracks them across frames, and evaluates performance using standard metrics. The package solves the problem of developing and benchmarking perception algorithms in a realistic, reproducible environment.
Autonomous driving researchers, robotics engineers, and computer vision developers working on perception systems who need a modular pipeline for object detection and tracking with KITTI dataset compatibility.
Developers choose this package because it provides a ready-to-use, modular ROS implementation of a complete perception pipeline with built-in evaluation against the standard KITTI benchmark. It offers transparency into each processing stage and allows easy experimentation with different algorithms while maintaining compatibility with widely-used datasets and metrics.
ROS package for the Perception (Sensor Processing, Detection, Tracking and Evaluation) of the KITTI Vision Benchmark Suite
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Allows step-by-step execution of sensor processing, detection, and tracking via separate ROS launch files, enabling easy experimentation with individual components as highlighted in the Philosophy section.
Includes built-in Python scripts to calculate standard MOTA, MOTP, MODA, and MODP metrics, providing direct benchmarking against KITTI Vision Benchmark standards, with demo results shown in the README.
Synchronizes Velodyne point cloud data with camera images for enhanced environmental perception, as described in the Key Features, leveraging ROS for integrated data handling.
Incorporates preprocessed semantic segmented images to improve detection reliability, ensuring CPU-friendly performance without real-time DNN inference, though it requires pre-generated data.
Requires a strict folder structure, manual data downloading from Google Drive, and troubleshooting issues like hardcoding home directory paths, as noted in the Troubleshooting section.
Primarily tailored to the KITTI dataset; adapting to other datasets like nuScenes would need significant modification, as the pipeline depends on KITTI-specific data formats and preprocessing.
Demo runs at 20% speed with intentional delays for RViz boot-up, making it unsuitable for live applications or real-time testing without code changes, per the Usage parameters.