No abstract

For self-driving systems or autonomous vehicles (AVs), accurate lane-level localization is a important for performing complex driving maneuvers. Classical GNSS-based methods are usually not accurate enough to have lane-level localization to support the AV’s maneuvers. LiDAR-based localization can provide accurate localization. However, the price of LiDARs is still one of the big issues preventing this kind of solution from becoming wide-spread commodity. Therefore, in this work, we propose a low-cost solution for lane-level localization using a vision-based system and a low-cost GPS to achieve high precision lane-level localization. Experiments in real-world and real-time demonstrate that the proposed method achieves good lane-level localization accuracy, outperforming solutions based on only GPS.

Autonomous driving systems must have the ability to monitor the kinematic behaviour of multiple obsta-cles. Therefore, 3D multi-object tracking (3D-MOT) is one of the crucial modules in autonomous driving to detect the presence of potential hazard movements such as human operated vehicles and pedestrians. In this work, we present a novel online 3D multi-tracking system that uses the Aggre-gated Euclidean Distances (AED) in data association module instead of using Intersection over Union (IoU) as a new metric. AED is used in order to obtain the relationship between predicted tracks and current object detections. There are several benefits from using AED in data association module. Firstly, it can reduce the system's complexity so that the execution time can be significantly reduced (as calculating Euclidean distances is much faster than obtaining 3D-IoU). Secondly, AED can provide distance measurement even when there is no overlaps between the predicted tracks and the current detections, while 3D-IoU produces zeros for non-overlapping cases. To demonstrate the validity of our proposed method, we performed extensive experiments on KITTI multi-tracking benchmark and nuScenes validation datasets. The experimental results are compared against the open-sourced state of the art 3D MOTs such as AB3DMOT, FANTrack, and mmMOT. Our method clearly outperforms the AB3DMOT baseline method and other methods in terms of accuracy and/or processing speed.

No abstract