Project Description:
The data set was recorded on January 2022 on the VTTI test track (Surface Street). This project developed a radar and LiDAR object tracking fusion technique for use on a 1/5th scaled vehicle. To address this and evaluate object tracking accuracy, raw sensor data from a Continental radar, Velodyne LiDAR, and an ArduSimple DGPS board was collected for both a full-sized vehicle and a small scale vehicle. The two vehicles were the only moving objects in the environment and performed three types of maneuvers within the FOV of the two static sensors. The first maneuver consisted of driving straight away and then straight back toward the sensors in the middle of their FOVs; the second consisted of driving away from and then back toward the sensors diagonally across their FOV; the third consisted of performing small s-turns when driving away from and then toward the sensors. The distance and speed of the maneuvers is scaled accordingly between the full scale and small scale vehicle. Data collection used ROS recordings and ROS messages to save the real time messages from the radar, LiDAR, and DGPS for later playback and analysis. The raw data was later used to identify and track the vehicle using only the radar, only the LiDAR, and then a fusion technique with the radar and LiDAR. The accuracy of each method was calculated using the DGPS as a ground truth, and metrics to compare the increase in accuracy using the fusion method was developed.
Data Scope:
The dataset includes 6 ROS recordings corresponding to one of the vehicle maneuvers of the small scale or full scale vehicle: 3 from the full scale vehicle trials and three from the small scale vehicle trials (Table 1, see Data Specification document below). The ROS bag recordings consist of time series data of the three main sensors (radar, LiDAR, and DGPS). The radar messages are stored in custom ARS430 ROS message types and contain radar point information including position, relative velocity, and other characteristics such as SNR, RCS, and false detection probability. LiDAR messages are stored in the common point cloud 2 messages that provide x, y, and z position and the intensity of each point within the point cloud. The DPGS board provides position and pose information in common ROS messages (geometry_msgs/Pose) in Cartesian coordinates relative to the static radar and LiDAR sensors.
Each recording is between approximately 70-200 MB that correspond to trial duration times of approximately 30 seconds.
Data Specification:
Please see Data Specification document below.
