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
Vehicle Detection in Aerial Images Based on 3D Depth Maps and Deep Neural Networks
Blekinge Institute of Technology, Faculty of Engineering, Department of Mathematics and Natural Sciences. (Systems Engineering)ORCID iD: 0000-0002-6834-5676
Blekinge Institute of Technology, Faculty of Engineering, Department of Mathematics and Natural Sciences.ORCID iD: 0000-0003-3707-2780
Blekinge Institute of Technology, Faculty of Engineering, Department of Mathematics and Natural Sciences.ORCID iD: 0000-0002-6643-312X
2021 (English)In: IEEE Access, E-ISSN 2169-3536, Vol. 9, p. 8381-8391Article in journal (Refereed) Published
Abstract [en]

Object detection in aerial images, particularly of vehicles, is highly important in remote sensing applications including traffic management, urban planning, parking space utilization, surveillance, and search and rescue. In this paper, we investigate the ability of three-dimensional (3D) feature maps to improve the performance of deep neural network (DNN) for vehicle detection. First, we propose a DNN based on YOLOv3 with various base networks, including DarkNet-53, SqueezeNet, MobileNet-v2, and DenseNet-201. We assessed the base networks and their performance in combination with YOLOv3 on efficiency, processing time, and the memory that each architecture required. In the second part, 3D depth maps were generated using pairs of aerial images and their parallax displacement. Next, a fully connected neural network (fcNN) was trained on 3D feature maps of trucks, semi-trailers and trailers. A cascade of these networks was then proposed to detect vehicles in aerial images. Upon the DNN detecting a region, coordinates and confidence levels were used to extract the corresponding 3D features. The fcNN used 3D features as the input to improve the DNN performance. The data set used in this work was acquired from numerous flights of an unmanned aerial vehicle (UAV) across two industrial harbors over two years. The experimental results show that 3D features improved the precision of DNNs from 88.23 % to 96.43 % and from 97.10 % to 100 % when using DNN confidence thresholds of 0.01 and 0.05, respectively. Accordingly, the proposed system was able to successfully remove 72.22 % to 100 % of false positives from the DNN outputs. These results indicate the importance of 3D features utilization to improve object detection in aerial images for future research. CCBY

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2021. Vol. 9, p. 8381-8391
Keywords [en]
Convolutional neural networks, 3D depth maps, Object detection, Aerial images
National Category
Computer Vision and Robotics (Autonomous Systems)
Identifiers
URN: urn:nbn:se:bth-20923DOI: 10.1109/ACCESS.2021.3049741ISI: 000608205500001Scopus ID: 2-s2.0-85099218070OAI: oai:DiVA.org:bth-20923DiVA, id: diva2:1517909
Note

open access

Available from: 2021-01-14 Created: 2021-01-14 Last updated: 2021-02-04Bibliographically approved
In thesis
1. Computer Vision for Traffic Surveillance Systems: Methods and Applications
Open this publication in new window or tab >>Computer Vision for Traffic Surveillance Systems: Methods and Applications
2021 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Computer vision solutions play a significant role in intelligent transportation systems (ITS) by improving traffic flow, safety and management. In addition, they feature prominently in autonomous vehicles and their future development. The main advantages of vision-based systems are their flexibility, coverage and accessibility. Moreover, computational power and recent algorithmic advances have increased the promise of computer vision solutions and broadened their implementation. However, computational complexity, reliability and efficiency remain among the challenges facing vision-based systems.

Most traffic surveillance systems in ITS comprise three major criteria: vehicle detection, tracking and classification. In this thesis, computer vision systems are introduced to accomplish goals corresponding to these three criteria: 1) to detect the changed regions of an industrial harbour's parking lot using aerial images, 2) to estimate the speed of the vehicles on the road using a stationary roadside camera and 3) to classify vehicles using a stationary roadside camera and aerial images.

The first part of this thesis discusses change detection in aerial images, which is the core of many remote sensing applications. The aerial images were taken over an industrial harbour using unmanned aerial vehicles on different days and under various circumstances. This thesis presents two approaches to detecting changed regions: a local pattern descriptor and three-dimensional feature maps. These methods are robust to varying illumination and shadows. Later, the introduced 3D feature map generation model was employed for vehicle detection in aerial images.

The second part of this thesis deals with vehicle speed estimation using roadside cameras. Information regarding the flow, speed and number of vehicles is essential for traffic surveillance systems. In this thesis, two vision-based vehicle speed estimation approaches are proposed. These analytical models consider the measurement uncertainties related to the camera sampling time. The main contribution of these models is to estimate a speed probability density function for every vehicle. Later, the speed estimation model was utilised for vehicle classification using a roadside camera.

Finally, in the third part, two vehicle classification models are proposed for roadside and aerial images. The first model utilises the proposed speed estimation method to extract the speed of the passing vehicles. Then, we used a fuzzy c-means algorithm to classify vehicles using their speeds and dimension features. The results show that vehicle speed is a useful feature for distinguishing different categories of vehicles. The second model employs deep neural networks to detect and classify heavy vehicles in aerial images. In addition, the proposed 3D feature generation model was utilised to improve the performance of the deep neural network. The experimental results show that 3D feature information can significantly reduce false positives in the deep learning model's output.

This thesis comprises two chapters: Introduction, and Publications. In the introduction section, we discuss the motivation for computer vision solutions and their importance. Furthermore, the concepts and algorithms used to construct the proposed methods are explained. The second chapter presents the included publications.

Place, publisher, year, edition, pages
Karlshamn: Blekinge Tekniska Högskola, 2021. p. 149
Series
Blekinge Institute of Technology Doctoral Dissertation Series, ISSN 1653-2090 ; 1
Keywords
Intelligent transportation systems, ITS, Computer visions systems
National Category
Engineering and Technology Electrical Engineering, Electronic Engineering, Information Engineering Signal Processing Computer Vision and Robotics (Autonomous Systems)
Research subject
Systems Engineering
Identifiers
urn:nbn:se:bth-20924 (URN)978-91-7295-416-8 (ISBN)
Public defence
2021-03-03, Zoom, 08:30 (English)
Opponent
Supervisors
Available from: 2021-01-15 Created: 2021-01-14 Last updated: 2022-02-18Bibliographically approved

Open Access in DiVA

fulltext(12653 kB)564 downloads
File information
File name FULLTEXT01.pdfFile size 12653 kBChecksum SHA-512
272b8a67eb1749bfd92508fb33e311794617362fea53638c325861cb34060c9b54007df7783fdb83abba5a5c9c81ccc64465303b30cca6c6e9f5882b26f3564a
Type fulltextMimetype application/pdf

Other links

Publisher's full textScopus

Authority records

Javadi, SalehDahl, MattiasPettersson, Mats

Search in DiVA

By author/editor
Javadi, SalehDahl, MattiasPettersson, Mats
By organisation
Department of Mathematics and Natural Sciences
In the same journal
IEEE Access
Computer Vision and Robotics (Autonomous Systems)

Search outside of DiVA

GoogleGoogle Scholar
Total: 566 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 454 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