FA3D: Fast and Accurate 3D Object Detection

Hamed Habibi Aghdam; Selameab Demilew; Robert Laganiére; Emil Petriu

doi:10.1007/978-3-030-64556-4_31

FA3D: Fast and Accurate 3D Object Detection

Hamed Habibi Aghdam, Selameab Demilew^*, Robert Laganiére, Emil Petriu

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Fast and accurate detection of objects, in 3D, is one of the critical components in an advanced driver assistance system. In this paper, we aim to develop an accurate 3D object detector that runs in near real-time on low-end embedded systems. We propose an efficient framework that converts raw point cloud into a 3D occupancy cuboid and detects cars using a deep convolutional neural network. Even though the complexity of our proposed model is high, it runs at 7.27 FPS on a Jetson Xavier and at 57.83 FPS on a high-end workstation that is 18 % and 43 % faster than the fastest published method while having a comparable performance with state-of-the-art models on the KITTI dataset. We conduct a comprehensive error analysis on our model and show that two quantities are the principal sources of error among nine predicted attributes. Our source code is available at https://github.com/Selameab/FA3D. © 2020, Springer Nature Switzerland AG.

Original language	English
Title of host publication	Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings
Editors	George Bebis, Zhaozheng Yin, Edward Kim, Jan Bender, Kartic Subr, Bum Chul Kwon, Jian Zhao, Denis Kalkofen, George Baciu
Pages	397-409
Number of pages	13
DOIs	https://doi.org/10.1007/978-3-030-64556-4_31
Publication status	Published - 7 Dec 2020
Externally published	Yes
Event	15th International Symposium on Visual Computing: ISVC 2020 - San Diego, United States Duration: 5 Oct 2020 → 7 Oct 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12509 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	15th International Symposium on Visual Computing
Country	United States
City	San Diego
Period	5/10/20 → 7/10/20

Keywords

3D object detection
Deep neural networks
Smart and autonomous vehicles

ASJC Scopus subject areas

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-030-64556-4_31

Link to publication in Scopus

Fingerprint Dive into the research topics of 'FA3D: Fast and Accurate 3D Object Detection'. Together they form a unique fingerprint.

Cite this

Habibi Aghdam, H., Demilew, S., Laganiére, R., & Petriu, E. (2020). FA3D: Fast and Accurate 3D Object Detection. In G. Bebis, Z. Yin, E. Kim, J. Bender, K. Subr, B. C. Kwon, J. Zhao, D. Kalkofen, ... G. Baciu (Eds.), Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings (pp. 397-409). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12509 LNCS). https://doi.org/10.1007/978-3-030-64556-4_31

FA3D: Fast and Accurate 3D Object Detection. / Habibi Aghdam, Hamed; Demilew, Selameab; Laganiére, Robert; Petriu, Emil.

Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings. ed. / George Bebis; Zhaozheng Yin; Edward Kim; Jan Bender; Kartic Subr; Bum Chul Kwon; Jian Zhao; Denis Kalkofen; George Baciu. 2020. p. 397-409 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12509 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Habibi Aghdam, H, Demilew, S, Laganiére, R & Petriu, E 2020, FA3D: Fast and Accurate 3D Object Detection. in G Bebis, Z Yin, E Kim, J Bender, K Subr, BC Kwon, J Zhao, D Kalkofen & G Baciu (eds), Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12509 LNCS, pp. 397-409, 15th International Symposium on Visual Computing, San Diego, United States, 5/10/20. https://doi.org/10.1007/978-3-030-64556-4_31

Habibi Aghdam H, Demilew S, Laganiére R, Petriu E. FA3D: Fast and Accurate 3D Object Detection. In Bebis G, Yin Z, Kim E, Bender J, Subr K, Kwon BC, Zhao J, Kalkofen D, Baciu G, editors, Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings. 2020. p. 397-409. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-64556-4_31

Habibi Aghdam, Hamed ; Demilew, Selameab ; Laganiére, Robert ; Petriu, Emil. / FA3D: Fast and Accurate 3D Object Detection. Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings. editor / George Bebis ; Zhaozheng Yin ; Edward Kim ; Jan Bender ; Kartic Subr ; Bum Chul Kwon ; Jian Zhao ; Denis Kalkofen ; George Baciu. 2020. pp. 397-409 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{fc11d686a61c4c91958a5b62c6a5e5f2,

title = "FA3D: Fast and Accurate 3D Object Detection",

abstract = "Fast and accurate detection of objects, in 3D, is one of the critical components in an advanced driver assistance system. In this paper, we aim to develop an accurate 3D object detector that runs in near real-time on low-end embedded systems. We propose an efficient framework that converts raw point cloud into a 3D occupancy cuboid and detects cars using a deep convolutional neural network. Even though the complexity of our proposed model is high, it runs at 7.27 FPS on a Jetson Xavier and at 57.83 FPS on a high-end workstation that is 18 {\%} and 43 {\%} faster than the fastest published method while having a comparable performance with state-of-the-art models on the KITTI dataset. We conduct a comprehensive error analysis on our model and show that two quantities are the principal sources of error among nine predicted attributes. Our source code is available at https://github.com/Selameab/FA3D. {\circledC} 2020, Springer Nature Switzerland AG.",

keywords = "3D object detection, Deep neural networks, Smart and autonomous vehicles",

author = "{Habibi Aghdam}, Hamed and Selameab Demilew and Robert Lagani{\'e}re and Emil Petriu",

year = "2020",

month = "12",

day = "7",

doi = "10.1007/978-3-030-64556-4_31",

language = "English",

isbn = "9783030645557",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

pages = "397--409",

editor = "George Bebis and Zhaozheng Yin and Edward Kim and Jan Bender and Kartic Subr and Kwon, {Bum Chul} and Jian Zhao and Denis Kalkofen and George Baciu",

booktitle = "Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings",

}

TY - GEN

T1 - FA3D: Fast and Accurate 3D Object Detection

AU - Habibi Aghdam, Hamed

AU - Demilew, Selameab

AU - Laganiére, Robert

AU - Petriu, Emil

PY - 2020/12/7

Y1 - 2020/12/7

N2 - Fast and accurate detection of objects, in 3D, is one of the critical components in an advanced driver assistance system. In this paper, we aim to develop an accurate 3D object detector that runs in near real-time on low-end embedded systems. We propose an efficient framework that converts raw point cloud into a 3D occupancy cuboid and detects cars using a deep convolutional neural network. Even though the complexity of our proposed model is high, it runs at 7.27 FPS on a Jetson Xavier and at 57.83 FPS on a high-end workstation that is 18 % and 43 % faster than the fastest published method while having a comparable performance with state-of-the-art models on the KITTI dataset. We conduct a comprehensive error analysis on our model and show that two quantities are the principal sources of error among nine predicted attributes. Our source code is available at https://github.com/Selameab/FA3D. © 2020, Springer Nature Switzerland AG.

AB - Fast and accurate detection of objects, in 3D, is one of the critical components in an advanced driver assistance system. In this paper, we aim to develop an accurate 3D object detector that runs in near real-time on low-end embedded systems. We propose an efficient framework that converts raw point cloud into a 3D occupancy cuboid and detects cars using a deep convolutional neural network. Even though the complexity of our proposed model is high, it runs at 7.27 FPS on a Jetson Xavier and at 57.83 FPS on a high-end workstation that is 18 % and 43 % faster than the fastest published method while having a comparable performance with state-of-the-art models on the KITTI dataset. We conduct a comprehensive error analysis on our model and show that two quantities are the principal sources of error among nine predicted attributes. Our source code is available at https://github.com/Selameab/FA3D. © 2020, Springer Nature Switzerland AG.

KW - 3D object detection

KW - Deep neural networks

KW - Smart and autonomous vehicles

UR - http://www.scopus.com/inward/record.url?scp=85098134042&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-64556-4_31

DO - 10.1007/978-3-030-64556-4_31

M3 - Conference contribution

SN - 9783030645557

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 397

EP - 409

BT - Advances in Visual Computing - 15th International Symposium, ISVC 2020, Proceedings

A2 - Bebis, George

A2 - Yin, Zhaozheng

A2 - Kim, Edward

A2 - Bender, Jan

A2 - Subr, Kartic

A2 - Kwon, Bum Chul

A2 - Zhao, Jian

A2 - Kalkofen, Denis

A2 - Baciu, George

ER -