This paper proposes a computationally fast and accurate framework to identify natively in 3D point clouds from LIDAR for autonomous driving. Speed is significant and detection is an indispensable component of safety. Many of the approaches are. However, most of them remain massive computation due to high dimensionality and densely of point clouds. We therefore investigate and carefully design an enhanced single-stage keypoint 3D multi-object detection framework to balance high accuracy and real-time efficiency. Experiments on the challenging KITTI dataset show that AEC3D outperforms other projection-based performances by a noteworthy margin. In addition, this is an embedded system-friendly approach, the inference times on TITAN Xp and Jetson AGX Xavier Developer Kit are 48 FPS and 14 FPS, respectively.
AEC3D: An Efficient and Compact Single Stage 3D Multiobject Detector for Autonomous Driving
IEEE Transactions on Intelligent Transportation Systems ; 23 , 12 ; 23422-23432
2022-12-01
4565299 byte
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Deep Learning for Model-Based Multiobject Tracking
IEEE | 2023
|Interactive 3D shape design — Multipatch and Multiobject
Emerald Group Publishing | 1973
|Multiobject tracking in video using a trisection paradigm
British Library Online Contents | 2007
|