Change search
Link to record
Permanent link

Direct link
Publications (10 of 104) Show all publications
Hallösta, S., Pettersson, M. & Dahl, M. (2024). Impact of Neural Network Architecture for Fingerprint Recognition. In: Akram Bennour, Ahmed Bouridane, Lotfi Chaari (Ed.), Intelligent Systems and Pattern Recognition: Third International Conference, ISPR 2023, Hammamet, Tunisia, May 11–13, 2023, Revised Selected Papers, Part I. Paper presented at 3rd International Conference on Intelligent Systems & Pattern Recognition, ISPR2023, Hammamet, 11/5 - 13/5 2023 (pp. 3-14). Springer, 1940
Open this publication in new window or tab >>Impact of Neural Network Architecture for Fingerprint Recognition
2024 (English)In: Intelligent Systems and Pattern Recognition: Third International Conference, ISPR 2023, Hammamet, Tunisia, May 11–13, 2023, Revised Selected Papers, Part I / [ed] Akram Bennour, Ahmed Bouridane, Lotfi Chaari, Springer, 2024, Vol. 1940, p. 3-14Conference paper, Published paper (Refereed)
Abstract [en]

This work investigates the impact of the neural networks architecture when performing fingerprint recognition. Three networks are studied; a Triplet network and two Siamese networks. They are evaluated on datasets with specified amounts of relative translation between fingerprints. The results show that the Siamese model based on contrastive loss performed best in all evaluated metrics. Moreover, the results indicate that the network with a categorical scheme performed inferior to the other models, especially in recognizing images with high confidence. The Equal Error Rate (EER) of the best model ranged between 4%−11% which was on average 6.5 percentage points lower than the categorical schemed model. When increasing the translation between images, the networks were predominantly affected once the translation reached a fourth of the image. Our work concludes that architectures designed to cluster data have an advantage when designing an authentication system based on neural networks.

Place, publisher, year, edition, pages
Springer, 2024
Series
Communications in Computer and Information Science, ISSN 1865-0929, E-ISSN 1865-0937 ; 1940
Keywords
Fingerprint recognition, Neural network architecture, Siamese network
National Category
Computer Vision and Robotics (Autonomous Systems)
Research subject
Systems Engineering
Identifiers
urn:nbn:se:bth-25604 (URN)10.1007/978-3-031-46335-8_1 (DOI)2-s2.0-85177185075 (Scopus ID)978-3-031-46334-1 (ISBN)978-3-031-46335-8 (ISBN)
Conference
3rd International Conference on Intelligent Systems & Pattern Recognition, ISPR2023, Hammamet, 11/5 - 13/5 2023
Available from: 2023-11-14 Created: 2023-11-14 Last updated: 2023-12-04Bibliographically approved
Fredriksson, H., Holmgren, J., Dahl, M. & Lövström, B. (2023). A Median-Based Misery Index for Travel Time Reliability. In: Elhadi Shakshuki (Ed.), Procedia Computer Science: . Paper presented at 14th International Conference on Ambient Systems, Networks and Technologies Networks, ANT 2023 and The 6th International Conference on Emerging Data and Industry 4.0, EDI40 2023, Leuven, 15 March through 17 March 2023 (pp. 162-169). Elsevier, 220
Open this publication in new window or tab >>A Median-Based Misery Index for Travel Time Reliability
2023 (English)In: Procedia Computer Science / [ed] Elhadi Shakshuki, Elsevier, 2023, Vol. 220, p. 162-169Conference paper, Published paper (Refereed)
Abstract [en]

Travel time reliability is vital for both road agencies and road users. Expected travel time reliability can be used by road agencies to assess the state of a transportation system, and by road users, to schedule their trips. Road network deficiencies, such as insufficient traffic flow capacity of a road segment or poor road design, have a negative impact on the reliability of travel times. Thus, to maintain robust and reliable travel times, the detection of road network deficiencies is vital. By continuously analyzing travel times and using appropriate travel time reliability measurements, it is possible to detect existing deficiencies or deficiencies that may eventually occur unless necessary actions are taken. In many cases, indices and measurements of travel time reliability are related to the distribution of the travel times, specifically the skewness and width of the distribution. The current paper introduces a median-based misery index for travel time reliability. The index is robust and handles travel times that follow a skewed distribution well. The index measures the relative difference between the slow travel speeds and the free-flow travel speed. The index is inspired by the median absolute deviation, and its primary application is to detect routes or road segments with potential road network deficiencies. To demonstrate the applicability of the index, we conducted an empirical case study using real travel speed data from the European route E4 in Sweden. The results from the empirical case study indicate that the index is capable of detecting road segments with slow travel speeds regardless of the travel speed distribution.

Place, publisher, year, edition, pages
Elsevier, 2023
Series
Procedia Computer Science, E-ISSN 1877-0509
Keywords
Travel time reliability, travel speed index, travel speed
National Category
Transport Systems and Logistics
Research subject
Mathematics and applications
Identifiers
urn:nbn:se:bth-24456 (URN)10.1016/j.procs.2023.03.023 (DOI)2-s2.0-85164538353 (Scopus ID)
Conference
14th International Conference on Ambient Systems, Networks and Technologies Networks, ANT 2023 and The 6th International Conference on Emerging Data and Industry 4.0, EDI40 2023, Leuven, 15 March through 17 March 2023
Funder
Swedish Transport Administration
Available from: 2023-04-20 Created: 2023-04-20 Last updated: 2023-08-07Bibliographically approved
Fredriksson, H., Dahl, M., Holmgren, J., Lövström, B., Irvenå, J. & Mårtensson, M. (2022). Förstudie – Datadriven analys av restider.
Open this publication in new window or tab >>Förstudie – Datadriven analys av restider
Show others...
2022 (Swedish)Report (Other academic)
Abstract [sv]

Det uppkopplade samhället möjliggör kontinuerlig insamling av trafikdata genom de moderna fordonens navigationssystem, så väl som genom mobiltelefoner och andra GPS-enheter. Den ökade tillgången till trafikdata ger möjligheter analysera att trafiksystemet ur flera olika perspektiv. En typ av trafikdata som kan extraheras ur GPS-data är fordonshastigheter. Genom att analysera hur fordonshastigheter förändras över tid och identifiera avvikelser från ett normaltillstånd så skulle det vara möjligt att upptäcka och förutsäga potentiella brister i trafiksystemet. Många gånger upptäcks brister i infrastrukturen i ett relativt sent skede vilket i sin tur kan innebära både omfattande och kostsamma åtgärder för att komma till rätta med problemen.

Syftet med förstudien har varit att utveckla och utvärdera metoder och modeller för att detektera brister och identifiera hastigheter som relativt avviker från normaltillståndet, dvs ett fokus på fordonshastigheter som framför allt är ovanligt låga. Utgångspunkten har också varit att finna lämplig metod för att modellera trafiksituationen med hjälp av uppmätta fordonshastigheter. Det vill säga metoder som syftar till att ur ett övergripande perspektiv beskriva normaltillståndet längs med de studerade vägsträckorna. Analyser av normaltillståndets förändring över tid öppnar upp möjligheten att detektera om brister relaterat till vägsträckors kapacitet och framkomlighet uppstått eller avgöra om normaltillståndet på en vägsträcka är stabilt eller förändras över tid.

Det är framför allt de relativt låga fordonshastigheterna som uppstår som blir en indikator på att en vägsträcka har brister. Därför föreslås en metod för att systematiskt identifiera och gruppera uppmätta fordonshastigheter i låga, normala och höga hastigheter. En utgångspunkt har varit robusthet och att möjliggöra jämförelser av hastigheter för olika vägsträckor med olika attribut som antal körfält och skyltad hastighet med varandra. Vi presenterar även ett nytt mått som beskriver hur gruppen med relativt låga hastigheter förhåller sig till friflödeshastigheten som till exempel den skyltade hastigheten. Syftet med måttet är att kvantifiera framkomligheten på en vägsträcka eller vägsegment. Existerande mått och indikatorer baseras idag på fordonshastigheter som spänner från låga till höga hastigheter. Vi har i denna kontext tagit fram ett mått som endast tar hänsyn till vad som anses vara låga hastigheter och friflödeshastighet.

Inom förstudien så har även en metod baserad på klusteranalys använts för de studerade vägsträckorna. Klusteranalys har i olika sammanhang visat sig effektivt för att kategorisera och detektera återkommande mönster i hastighetsprofiler. Syftet med klusteranalysen är att undersöka om det finns någon koppling mellan hastighetsprofiler som har liknande beteende och till exempel veckodag och tidpunkt. Genom klusteranalys skulle det vara möjligt att inte bara detektera vilka vägsträckor där det uppstår problem, utan det skulle även vara möjligt att prognostisera vid vilka veckodagar och tidpunkter där det finns risk att köer och andra problem kan uppstå.

Förstudien är begränsad till användning av fordonhastigheter som datakälla och de framtagna metoderna och modellerna visar att det finns potential att frikoppla sig från andra datakällor som till exempel fordonsflöden för att detektera brister eller avvikelser som skulle kunna indikera brister i transportsystemet.

Abstract [en]

Modern vehicles are to a large extent connected today, either directly by built-in navigation systems in the vehicles or indirectly by other devices such as mobile phones and GPS units. This enables the possibility to continuously collect traffic data in a cost-effective way. The increased access to detailed data allows practitioners and researchers to analyze the transportation system from various perspectives. The travel speed is a common descriptor of the traffic state, and it can be extracted from GPS data. By analyzing how the travel speed vary over time and detect anomalies among the measured travel speeds, it is possible to detect potential deficiencies in the transportation system, e.g., insufficient road capacity which may cause bottlenecks. Often, a weakness in the infrastructure is detected in a very late stage which means that extensive investments may be required to resolve the deficiency.

The purpose of the pilot study is to develop methods and models to detect deficiency in the transportation system and to identity travel speeds that deviates from the normal state, i.e., travel speeds that are considered as very low or very high with respect to the normal behavior. Thus, the starting point of the pilot study is to find appropriate ways to model the traffic state along the studied road segments by using measured travel speeds from a general point of view. Analysis of the traffic state allows the study of how the normal state of the road segments change of time to detect deficiency related to road capacity and road access which may occur if no changes are made, or to detect road segments where the normal state is unchanged.

Typically, slower travel speeds may be an indicator of that a deficiency along a road segment exists. Thus, we present a method to systematically partition measured travel speeds in low, normal, and high travel speeds. The method is robust and enable the possibility to compare different road segment with different attributes, such as number of lanes and free-flow travel speed, with each other. Furthermore, we present a new measurement to describe how the low travel speeds relates to the free flow travel speed, e.g., the speed limit. Existing measurements and indicators used today utilize travel speeds which range from low to high. Our proposed measurement uses low travel speed and free flow travel speed exclusively and aims to quantify the accessibility and condition of a road segment.

The pilot study also includes an initial attempt to apply cluster analysis to detect recurrent patterns along the studied road segments. Cluster analysis is in several contexts an effective method to group time series to detect recurrent patterns among the speed profiles. The purpose of using cluster analysis is to evaluate if speed profiles with similar behavior is related to, for instance, weekday or time of the day. Thus, cluster analysis may be used to detect road segments with recurring low travel speeds, and potentially be used to forecast when congestion or queues may occur.

The pilot study is mainly limited to travel speed data. The proposed methods and models show that it is possibly to solely use travel speed data to detect deficiencies in the transportation system. In particular, the pilot study shows the potential to detect deficiencies in the transportation system without additional data sources such as link flow data.

National Category
Transport Systems and Logistics
Research subject
Mathematics and applications
Identifiers
urn:nbn:se:bth-25030 (URN)
Funder
Swedish Transport Administration
Available from: 2023-06-28 Created: 2023-06-28 Last updated: 2023-06-28Bibliographically approved
Rameez, M., Pettersson, M. & Dahl, M. (2022). Interference Compression and Mitigation for Automotive FMCW Radar Systems. IEEE Sensors Journal, 22(20), 19739-19749
Open this publication in new window or tab >>Interference Compression and Mitigation for Automotive FMCW Radar Systems
2022 (English)In: IEEE Sensors Journal, ISSN 1530-437X, E-ISSN 1558-1748, Vol. 22, no 20, p. 19739-19749Article in journal (Refereed) Published
Abstract [en]

Millimeter-wave (mm-wave) frequency-modu- lated continuous wave (FMCW) radars are increasingly being deployed for scenario perception in various applications. It is expected that the mutual interference between such radars will soon become a significant problem. Therefore, to maintain the reliability of the radar measurements, there must be procedures in place to mitigate this interference. This article proposes a novel interference mitigation technique that utilizes the pulse compression principle for interference compression and mitigation. The interference in the received time-domain signal is compressed using an estimated matched filter. Afterward, the compressed interference is discarded, and the signal is repaired in the pulse-compressed domain using an autoregressive (AR) model. Since the interference spans fewer samples after compression, the signal can be restored more accurately in the compressed domain. Real outdoor measurements show that the interference is effectively suppressed down to the noise floor using the proposed scheme. A signal to interference and noise ratio (SINR) gain of approximately 14 dB was achieved in the experimental data, supporting this study. Moreover, the results indicate that this method is also applicable to situations where multiple interference sources are present.

Place, publisher, year, edition, pages
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC, 2022
Keywords
Interference, Radar, Chirp, Sensors, Time-domain analysis, Time-frequency analysis, Radar detection, Automotive radar, interference cancellation, millimeter wave (mm-wave) radar, pulse compression, radar signal processing, signal reconstruction
National Category
Signal Processing
Identifiers
urn:nbn:se:bth-23838 (URN)10.1109/JSEN.2022.3204505 (DOI)000870341900066 ()
Note

open access

This work was supported in part by the Netport Science Park and in part by the Municipality of Karlshamn.

Available from: 2022-11-04 Created: 2022-11-04 Last updated: 2023-02-08Bibliographically approved
Rameez, M., Dahl, M. & Pettersson, M. (2021). Autoregressive Model-Based Signal Reconstruction for Automotive Radar Interference Mitigation. IEEE Sensors Journal (5), 6575-6586
Open this publication in new window or tab >>Autoregressive Model-Based Signal Reconstruction for Automotive Radar Interference Mitigation
2021 (English)In: IEEE Sensors Journal, ISSN 1530-437X, E-ISSN 1558-1748, no 5, p. 6575-6586Article in journal (Refereed) Published
Abstract [en]

Automotive radars have become an important part of sensing systems in vehicles and other traffic applications due to their accuracy, compact design, and robustness under severe light and weather conditions. The increased use of radars in various traffic applications has given rise to the problem of mutual interference, which needs to be mitigated. In this paper, we investigate interference mitigation in chirp sequence (CS) automotive radars via signal reconstruction based on autoregressive (AR) models in fast-and slow-time. The interference is mitigated by replacing the disturbed baseband signal samples with samples predicted using the estimated AR models. Measurements from 77 GHz frequency modulated continuous wave (FMCW) static and moving radars are used to evaluate the signal reconstruction performance in terms of the signal-to-interference-plus-noise ratio (SINR), peak side-lobe level (PSLL), and mean squared error (MSE). The results show that the interference is suppressed down to the general noise floor, leading to an improvement in the SINR. Additionally, enhanced side-lobe suppression is achieved via AR signal reconstruction, which is compared to a commonly used inverse-cosine method. Furthermore, the paper notes that the slow-time signal reconstruction can be more beneficial for interference suppression in certain scenarios. CCBY

Place, publisher, year, edition, pages
IEEE, 2021
Keywords
Automotive radar, Autoregressive (AR) modelling, Chirp Sequence (CS), Frequency Modulated Continuous Wave (FMCW), interference mitigation, signal reconstruction
National Category
Electrical Engineering, Electronic Engineering, Information Engineering Signal Processing
Identifiers
urn:nbn:se:bth-19375 (URN)10.1109/JSEN.2020.3042061 (DOI)000616329300107 ()2-s2.0-85097401154 (Scopus ID)
Note

open access

Available from: 2020-04-07 Created: 2020-04-07 Last updated: 2023-02-08Bibliographically approved
Fredriksson, H., Dahl, M., Lövström, B., Holmgren, J. & Lennerstad, H. (2021). Modeling of road traffic flows in the neighboring regions. In: Shakshuki E., Yasar A. (Ed.), Procedia Computer Science: . Paper presented at The 12th International Conference on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN), Leuven, Belgium, November 1-4, 2021 (pp. 43-50). Elsevier
Open this publication in new window or tab >>Modeling of road traffic flows in the neighboring regions
Show others...
2021 (English)In: Procedia Computer Science / [ed] Shakshuki E., Yasar A., Elsevier, 2021, p. 43-50Conference paper, Published paper (Refereed)
Abstract [en]

Traffic flows play a very important role in transportation engineering. In particular, link flows are a source of information about the traffic state, which is usually available from the authorities that manage road networks. Link flows are commonly used in both short-term and long-term planning models for operation and maintenance, and to forecast the future needs of transportation infrastructure. In this paper, we propose a model to study how traffic flow in one location can be expected to reflect the traffic flow in a nearby region. The statistical basis of the model is derived from link flows to find estimates of the distribution of traffic flows in junctions. The model is evaluated in a numerical study, which uses real link flow data from a transportation network in southern Sweden. The results indicate that the model may be useful for studying how large departing flows from a node reflect the link flows in a neighboring geographic region. 

Place, publisher, year, edition, pages
Elsevier, 2021
Series
Procedia Computer Science, E-ISSN 1877-0509 ; 198
Keywords
link flows, traffic volumes, flow distribution, flow estimation, transportation network
National Category
Transport Systems and Logistics
Research subject
Mathematics and applications
Identifiers
urn:nbn:se:bth-22071 (URN)10.1016/j.procs.2021.12.209 (DOI)2-s2.0-85124595881 (Scopus ID)
Conference
The 12th International Conference on Emerging Ubiquitous Systems and Pervasive Networks (EUSPN), Leuven, Belgium, November 1-4, 2021
Note

open access

Available from: 2021-09-01 Created: 2021-09-01 Last updated: 2022-12-02Bibliographically approved
Holmgren, J., Fredriksson, H. & Dahl, M. (2021). On the use of active mobile and stationary devices for detailed traffic data collection: A simulation-based evaluation. International Journal of Traffic and Transportation Management, 3(1), 1-9
Open this publication in new window or tab >>On the use of active mobile and stationary devices for detailed traffic data collection: A simulation-based evaluation
2021 (English)In: International Journal of Traffic and Transportation Management, ISSN 2371-5782, Vol. 3, no 1, p. 1-9Article in journal (Refereed) Published
Abstract [en]

The process of collecting traffic data is a key component to evaluate the current state of a transportation network and to analyze movements of vehicles. In this paper, we argue that both active stationary and mobile measurement devices should be taken into account for high-quality traffic data with sufficient geographic coverage. Stationary devices are able to collect data over time at certain locations in the network and mobile devices are able to gather data over large geographic regions. Hence, the two types of measurement devices have complementary properties and should be used in conjunction with each other in the data collection process. To evaluate the complementary characteristics of stationary and mobile devices for traffic data collection, we present a traffic simulation model, which we use to study the share of successfully identified vehicles when using both types of devices with varying identification rate. The results from our simulation study, using freight transport in southern Sweden, shows that the share of successfully identified vehicles can be significantly improved by using both stationary and mobile measurement devices.

Place, publisher, year, edition, pages
The International Association for Sharing Knowledge and Sustainability (IASKS), 2021
Keywords
Traffic data collection, stationary devices, mobile devices, traffic simulation
National Category
Transport Systems and Logistics
Research subject
Mathematics and applications
Identifiers
urn:nbn:se:bth-21315 (URN)10.5383/JTTM.03.01.001 (DOI)
Available from: 2021-04-03 Created: 2021-04-03 Last updated: 2023-01-06Bibliographically approved
Fredriksson, H., Dahl, M. & Holmgren, J. (2021). Optimal Allocation of Charging Stations for Electric Vehicles Using Probabilistic Route Selection. Computing and informatics, 40(2), 408-427
Open this publication in new window or tab >>Optimal Allocation of Charging Stations for Electric Vehicles Using Probabilistic Route Selection
2021 (English)In: Computing and informatics, ISSN 1335-9150, Vol. 40, no 2, p. 408-427Article in journal (Refereed) Published
Abstract [en]

Electric vehicles (EVs) are environmentally friendly and are considered to be a promising approach toward a green transportation infrastructure with lower greenhouse gas emissions. However, the limited driving range of EVs demands a strategic allocation of charging facilities, hence providing recharging opportunities that help reduce EV owners' anxiety about their vehicles' range. In this paper, we study a set covering method where self-avoiding walks are utilized to find the most significant locations for charging stations. In the corresponding optimization problem, we derive a lower bound of the number of charging stations in a transportation network to obtain full coverage of the most probable routes. The proposed method is applied to a transportation network of the southern part of Sweden.

Place, publisher, year, edition, pages
SLOVAK ACAD SCIENCES INST INFORMATICS, 2021
Keywords
Charging stations, electric vehicle, transportation network, optimal placement, self-avoiding random walk
National Category
Transport Systems and Logistics
Research subject
Mathematics and applications
Identifiers
urn:nbn:se:bth-22203 (URN)10.31577/cai_2021_2_408 (DOI)000718900400008 ()2-s2.0-85118310218 (Scopus ID)
Note

open access

Available from: 2021-10-13 Created: 2021-10-13 Last updated: 2021-12-03Bibliographically approved
Rameez, M., Javadi, S., Dahl, M. & Pettersson, M. (2021). Signal Reconstruction Using Bi-LSTM for Automotive Radar Interference Mitigation. In: 2021 18th European Radar Conference, EuRAD 2021: . Paper presented at 18th European Radar Conference, EuRAD 2021, London, 5 April 2022 through 7 April 2022 (pp. 74-77). Institute of Electrical and Electronics Engineers (IEEE)
Open this publication in new window or tab >>Signal Reconstruction Using Bi-LSTM for Automotive Radar Interference Mitigation
2021 (English)In: 2021 18th European Radar Conference, EuRAD 2021, Institute of Electrical and Electronics Engineers (IEEE), 2021, p. 74-77Conference paper, Published paper (Refereed)
Abstract [en]

Automotive radar has emerged as an important sensor for environmental perception in modern vehicles. A rapid increase in the number of radars present in traffic operating at unregulated frequencies has given rise to a mutual interference problem. In order for radar-based systems to function reliably, such interference must be mitigated. In this paper, this problem is addressed with a bidirectional long short-term memory (Bi-LSTM) network as a deep learning approach. Using the Bi-LSTM network, we reconstruct the intermediate frequency (IF) signal and recover samples lost to interference. The proposed signal reconstruction method is evaluated via real measurement data. The proposed Bi-LSTM network provides a better performance than an autoregressive model-based signal reconstruction method. © 2022 European Microwave Association (EuMA).

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2021
Keywords
FMCW radar, Interference suppression, Radar signal processing, Recurrent neural networks, Signal reconstruction, Automotive radar, Continuous wave radar, Frequency modulation, Long short-term memory, Radar equipment, Radar interference, Environmental perceptions, Interference mitigation, Interference problems, Learning approach, Memory network, Mutual interference, Reconstruction method, Signals reconstruction
National Category
Signal Processing
Identifiers
urn:nbn:se:bth-23510 (URN)10.23919/EuRAD50154.2022.9784516 (DOI)000838709300018 ()2-s2.0-85133132279 (Scopus ID)9782874870651 (ISBN)
Conference
18th European Radar Conference, EuRAD 2021, London, 5 April 2022 through 7 April 2022
Available from: 2022-08-11 Created: 2022-08-11 Last updated: 2023-02-08Bibliographically approved
Javadi, S., Dahl, M. & Pettersson, M. (2021). Vehicle Detection in Aerial Images Based on 3D Depth Maps and Deep Neural Networks. IEEE Access, 9, 8381-8391
Open this publication in new window or tab >>Vehicle Detection in Aerial Images Based on 3D Depth Maps and Deep Neural Networks
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
Keywords
Convolutional neural networks, 3D depth maps, Object detection, Aerial images
National Category
Computer Vision and Robotics (Autonomous Systems)
Identifiers
urn:nbn:se:bth-20923 (URN)10.1109/ACCESS.2021.3049741 (DOI)000608205500001 ()2-s2.0-85099218070 (Scopus ID)
Note

open access

Available from: 2021-01-14 Created: 2021-01-14 Last updated: 2021-02-04Bibliographically approved
Organisations
Identifiers
ORCID iD: ORCID iD iconorcid.org/0000-0003-3707-2780

Search in DiVA

Show all publications