CMHT autonomous dataset: A multi-sensor dataset including radar and IR for autonomous driving.

Area of Science:

• Robotics
• Computer Vision
• Sensor Technology

Background:

• Standardized datasets are crucial for advancing autonomous driving (AD) algorithms.
• Increasing sensor diversity necessitates datasets with aligned, multi-modal sensor data.
• Existing datasets may lack comprehensive sensor suites or diverse environmental conditions.

Purpose of the Study:

• To present a novel, comprehensive driving dataset for autonomous driving research.
• To facilitate the development and evaluation of perception and sensor fusion algorithms.
• To provide synchronized, multi-sensor data covering varied driving scenarios.

Main Methods:

• Recorded over 9000 frames of data at 10-20Hz using a complete sensor suite.
• Integrated Velodyne LiDAR, GPS/IMU, mm-wave radar, and color/infrared cameras.
• Captured data in diverse conditions including poor weather (rain/night) and varied traffic (highways/cities).

Main Results:

• The dataset includes fully synchronized data and raw ROS2 bags.
• Provides 3D tracklet labels for individual objects within the captured scenes.
• Details technical specifications of the driving platform, data format, and associated utilities.

Conclusions:

• The presented dataset offers a valuable resource for the autonomous driving community.
• Enables robust research in sensor fusion, perception, and algorithm development.
• Supports the creation of more reliable and adaptable autonomous driving systems.