L3D-OTVE: LiDAR-Based 3-D Object Tracking and Velocity Estimation Using LiDAR Odometry

Object tracking and velocity estimation are essential aspects of autonomous vehicles (AVs). AVs take actions based on tracking and estimating the velocity of obstacles moving around. Estimating the absolute velocity of obstacles poses challenges for optical sensors, such as LiDAR or cameras, in the absence of a global reference frame. Not much work has been done on estimating obstacle velocity using a LiDAR sensor. This letter proposes a novel method for tracking the velocity of obstacles using a LiDAR sensor, which involves transforming the current detection onto the previous axes of the LiDAR frame using LiDAR odometry, considering parameters, such as rotation and translation, between two consecutive frames. The estimated velocity is evaluated against ground truth data from the TiAND dataset. This dataset employs global navigation satellite system (GNSS) sensors equipped on a car acting as an obstacle, providing precise velocity measurements. The proposed algorithm demonstrates a root-mean-square deviation (RMSD) of 0.13 m/s, showcasing its accuracy in velocity estimation. The algorithm successfully tracks high-speed obstacles, reaching up to 100 km/h in highway data from the TiAND dataset and has also been tested on the KITTI dataset in real-time.