國立虎尾科技大學 |

Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement : = Accident Anticipation, Risky Agent Localization, and Trajectory Prediction.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement :/
Reminder of title:	Accident Anticipation, Risky Agent Localization, and Trajectory Prediction.
Author:	Karim, Muhammad Monjurul.
Description:	1 online resource (105 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
Contained By:	Dissertations Abstracts International85-02B.
Subject:	Civil engineering. -
Online resource:	click for full text (PQDT)
ISBN:	9798380141055

Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement : = Accident Anticipation, Risky Agent Localization, and Trajectory Prediction.
Karim, Muhammad Monjurul.

Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement :Accident Anticipation, Risky Agent Localization, and Trajectory Prediction. - 1 online resource (105 pages)

Source: Dissertations Abstracts International, Volume: 85-02, Section: B.

Thesis (Ph.D.)--State University of New York at Stony Brook, 2023.

Includes bibliographical references

Autonomous driving has gained significant attention recently, but it is expected that human drivers will continue to operate the majority of vehicles in the near future, with a mix of no and partial automation. In these mixed modes of traffic, the pressing issue of traffic safety and the associated risks of crashes persist, resulting in an annual fatality count of 35,000 and 1.8 million injuries in the United States, with an estimated economic loss of over $340 billion.To address this problem, an advanced accident-avoidance system is needed to avoid collisions and reduce crash severity. This involves several challenges to solve, including determining if an anomalous event is likely to occur in the near future, identifying traffic agents that will create the anomalous situation, and determining the current and future location of those risky traffic agents. In light to solving these challenges, this dissertation proposes a computer vision-based approach using deep learning models to develop such a system for traffic accident avoidance. Proposed approach includes the development of a human-inspired dynamic spatial-temporal attention network (DSTA) that effectively learns spatio-temporal feature relationships for future traffic accident anticipation in traffic scenes. I also use saliency maps to visually explain the prediction of the deep neural network for accidents. Then, I developed an attention-guided multistream feature fusion network (AM-Net) for localization of risky traffic agents from dashcam videos ahead of potential accidents. Finally, I introduced a novel holistic GRU encoder in an encoder-decoder deep neural network, enabling the anticipation of future traffic agent locations in anomalous driving videos.The dissertation research makes an original and novel contribution to the field of traffic safety, as demonstrated by the experimental evaluation. Proposed approach achieved state-of-the-art performance on a public dashcam video dataset, highlighting its potential impact on reducing traffic accidents, injuries, and fatalities. Overall, the dissertation research represents a significant step forward in enhancing roadway safety and has the potential to save countless lives.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798380141055Subjects--Topical Terms:

561339
Civil engineering.
Subjects--Index Terms:

Deep learningIndex Terms--Genre/Form:

554714
Electronic books.

Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement : = Accident Anticipation, Risky Agent Localization, and Trajectory Prediction.
LDR:03737ntm a22004217 4500 001 1148518
005 20240924103821.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798380141055
035 $a (MiAaPQ)AAI30494152
035 $a AAI30494152
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Karim, Muhammad Monjurul. $3 1474493
245 1 0 $a Deep Learning-Based Dashcam Video Analysis for Roadway Safety Enhancement : $b Accident Anticipation, Risky Agent Localization, and Trajectory Prediction.
264 0 $c 2023
300 $a 1 online resource (105 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
500 $a Advisor: Qin, Ruwen;Yin, Zhaozheng.
502 $a Thesis (Ph.D.)--State University of New York at Stony Brook, 2023.
504 $a Includes bibliographical references
520 $a Autonomous driving has gained significant attention recently, but it is expected that human drivers will continue to operate the majority of vehicles in the near future, with a mix of no and partial automation. In these mixed modes of traffic, the pressing issue of traffic safety and the associated risks of crashes persist, resulting in an annual fatality count of 35,000 and 1.8 million injuries in the United States, with an estimated economic loss of over $340 billion.To address this problem, an advanced accident-avoidance system is needed to avoid collisions and reduce crash severity. This involves several challenges to solve, including determining if an anomalous event is likely to occur in the near future, identifying traffic agents that will create the anomalous situation, and determining the current and future location of those risky traffic agents. In light to solving these challenges, this dissertation proposes a computer vision-based approach using deep learning models to develop such a system for traffic accident avoidance. Proposed approach includes the development of a human-inspired dynamic spatial-temporal attention network (DSTA) that effectively learns spatio-temporal feature relationships for future traffic accident anticipation in traffic scenes. I also use saliency maps to visually explain the prediction of the deep neural network for accidents. Then, I developed an attention-guided multistream feature fusion network (AM-Net) for localization of risky traffic agents from dashcam videos ahead of potential accidents. Finally, I introduced a novel holistic GRU encoder in an encoder-decoder deep neural network, enabling the anticipation of future traffic agent locations in anomalous driving videos.The dissertation research makes an original and novel contribution to the field of traffic safety, as demonstrated by the experimental evaluation. Proposed approach achieved state-of-the-art performance on a public dashcam video dataset, highlighting its potential impact on reducing traffic accidents, injuries, and fatalities. Overall, the dissertation research represents a significant step forward in enhancing roadway safety and has the potential to save countless lives.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Civil engineering. $3 561339
650 4 $a Transportation. $3 558117
650 4 $a Computer engineering. $3 569006
653 $a Deep learning
653 $a Autonomous driving
653 $a Traffic agents
653 $a Traffic accident
653 $a Anticipation
653 $a DSTA
655 7 $a Electronic books. $2 local $3 554714
690 $a 0543
690 $a 0709
690 $a 0464
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a State University of New York at Stony Brook. $b Civil Engineering. $3 1474494
773 0 $t Dissertations Abstracts International $g 85-02B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30494152 $z click for full text (PQDT)