umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Real-time DATMO Based on Deep 3D Point Cloud Features3D object detection in point cloud data
Umeå University, Faculty of Science and Technology, Department of Computing Science.
2019 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Perception for autonomous drive systems is the most essential task for safe and reliable driving. LiDAR sensors can be used for this and are vying for being crowned as an essentia lelement in this task.In this thesis, we present a novel real-time solution for detection and tracking of moving objects which utilizes deep learning based 3D object detection. On one hand, we present YOLO++, a 3D object detection network on point clouds only. A network that expands YOLOv3, the latest contribution to standard real-time object detector for three channel images. YOLO++ extracts the standard YOLO’s predictions plus an angle and a height from projected point clouds. Our unified architecture is fast. It processes images in 20 frames per second. Our experiments on the KITTI benchmark suite show that we achieve state-of-the-art efficiency but with a mediocre accuracy for car detection comparable to the result of Tiny-YOLOv3 on the COCO dataset.On the other hand, we present a multi-threaded object tracking solution that makes use o fthe detected objects by YOLO++. Each observation is associated to a thread with a novel concurrent data association process where each of the threads contain an Extended Kalman Filter that is used for predicting and estimating the object’s state over time. Futhermore, a LiDAR odometry algorithm is used to obtain absolute information about the movement, since the movement of objects are inherently relative to the sensor perceiving them. We obtain 33 state updates per second with an equal amount of threads to the number of cores inour main workstation.Even if the joint solution has not been tested on a system with enough computational power it is ready for deployment. We expect it to be runtime constrained by the slowest subsystem which happens to be the object detection system. This satisfies our real-time constraint of 10 frames per second of our final system by a large margin. Finally, we show that our system can take advantage of the predicted semantic information from the Kalman Filters in order to enhance the inference process in our YOLO++ architecture.

Place, publisher, year, edition, pages
2019. , p. 110
Series
Doctoral thesis / Umeå University, Department of Molecular Biology
Series
UMNAD ; 1202
National Category
Engineering and Technology
Identifiers
URN: urn:nbn:se:umu:diva-164718OAI: oai:DiVA.org:umu-164718DiVA, id: diva2:1366346
External cooperation
Volvo Cars
Educational program
Master of Science Programme in Computing Science and Engineering
Supervisors
Examiners
Available from: 2019-10-29 Created: 2019-10-29 Last updated: 2019-10-29Bibliographically approved

Open Access in DiVA

No full text in DiVA

By organisation
Department of Computing Science
Engineering and Technology

Search outside of DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric score

urn-nbn
Total: 36 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf