Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.3.mobmu-364
Taraka Rama Krishna Kanth Kannuri, Kirsnaragavan Arudpiragasam, Klaus Schwarz, Michael Hartmann, Reiner Creutzburg
Accident detection is one of the biggest challenges as there are various anomalies, occlusions, and objects in the image at different times. Therefore, this paper focuses on detecting traffic accidents through a combination of Object Tracking (OT) and image generation using GAN with variants such as skip connection, residual, and attention connection. The background removal techniques will be applied to reduce the background variation in the frame. Later, YOLO-R is used to detect objects, followed by DeepSort tracking of objects in the frame. Finally, the distance error metric and the adversarial error are determined using the Kalman filter and the GAN approach and help to decide accidents in videos.
{"title":"Generative adversarial networks (GANs) and object tracking (OT) for vehicle accident detection","authors":"Taraka Rama Krishna Kanth Kannuri, Kirsnaragavan Arudpiragasam, Klaus Schwarz, Michael Hartmann, Reiner Creutzburg","doi":"10.2352/ei.2023.35.3.mobmu-364","DOIUrl":"https://doi.org/10.2352/ei.2023.35.3.mobmu-364","url":null,"abstract":"Accident detection is one of the biggest challenges as there are various anomalies, occlusions, and objects in the image at different times. Therefore, this paper focuses on detecting traffic accidents through a combination of Object Tracking (OT) and image generation using GAN with variants such as skip connection, residual, and attention connection. The background removal techniques will be applied to reduce the background variation in the frame. Later, YOLO-R is used to detect objects, followed by DeepSort tracking of objects in the frame. Finally, the distance error metric and the adversarial error are determined using the Kalman filter and the GAN approach and help to decide accidents in videos.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"138 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135694714","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.16.avm-128
Shihao Shen, Louis Kerofsky, Senthil Yogamani
Estimating optical flow presents unique challenges in AV applications: large translational motion, wide variations in depth of important objects, strong lens distortion in commonly used fisheye cameras and rolling shutter artefacts in dynamic scenes. Even simple translational motion can produce complicated optical flow fields. Lack of ground truth data also creates a challenge. We evaluate recent optical flow methods on fisheye imagery found in AV applications. We explore various training techniques in challenging scenarios and domain adaptation for transferring models trained on synthetic data where ground truth is available to real-world data. We propose novel strategies that facilitate learning robust representations efficiently to address low-light degeneracies. Finally, we discuss the main challenges and open problems in this problem domain.
{"title":"Optical flow for autonomous driving: Applications, challenges and improvements","authors":"Shihao Shen, Louis Kerofsky, Senthil Yogamani","doi":"10.2352/ei.2023.35.16.avm-128","DOIUrl":"https://doi.org/10.2352/ei.2023.35.16.avm-128","url":null,"abstract":"Estimating optical flow presents unique challenges in AV applications: large translational motion, wide variations in depth of important objects, strong lens distortion in commonly used fisheye cameras and rolling shutter artefacts in dynamic scenes. Even simple translational motion can produce complicated optical flow fields. Lack of ground truth data also creates a challenge. We evaluate recent optical flow methods on fisheye imagery found in AV applications. We explore various training techniques in challenging scenarios and domain adaptation for transferring models trained on synthetic data where ground truth is available to real-world data. We propose novel strategies that facilitate learning robust representations efficiently to address low-light degeneracies. Finally, we discuss the main challenges and open problems in this problem domain.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"128 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135644696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.1.vda-394
Elias Neuman-Donihue, Michael Jarvis, Yuhao Zhu
In this paper, we introduce FastPoints, a state-of-the-art point cloud renderer for the Unity game development platform. Our program supports standard unprocessed point cloud formats with non-programmatic, drag-and-drop support, and creates an out-of-core data structure for large clouds without requiring an explicit preprocessing step; instead, the software renders a decimated point cloud immediately and constructs a shallow octree online, during which time the Unity editor remains fully interactive.
{"title":"FastPoints: A state-of-the-art point cloud renderer for Unity","authors":"Elias Neuman-Donihue, Michael Jarvis, Yuhao Zhu","doi":"10.2352/ei.2023.35.1.vda-394","DOIUrl":"https://doi.org/10.2352/ei.2023.35.1.vda-394","url":null,"abstract":"In this paper, we introduce FastPoints, a state-of-the-art point cloud renderer for the Unity game development platform. Our program supports standard unprocessed point cloud formats with non-programmatic, drag-and-drop support, and creates an out-of-core data structure for large clouds without requiring an explicit preprocessing step; instead, the software renders a decimated point cloud immediately and constructs a shallow octree online, during which time the Unity editor remains fully interactive.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135694179","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.3.mobmu-356
Valeria Vishnevskaya, Klaus Schwarz, Reiner Creutzburg
This paper presents a practical Open Source Intelligence (OSINT) use case for user similarity measurements with the use of open profile data from the Reddit social network. This PoC work combines the open data from Reddit and the part of the state-of-the-art BERT model. Using the PRAW Python library, the project fetches comments and posts of users. Then these texts are converted into a feature vector - representation of all user posts and comments. The main idea here is to create a comparable user's pair similarity score based on their comments and posts. For example, if we fix one user and calculate scores of all mutual pairs with other users, we will produce a total order on the set of all mutual pairs with that user. This total order can be described as a degree of written similarity with this chosen user. A set of "similar" users for one particular user can be used to recommend to the user interesting for him people. The similarity score also has a "transitive property": if $user_1$ is "similar" to $user_2$ and $user_2$ is similar to $user_3$ then inner properties of our model guarantees that $user_1$ and $user_3$ are pretty "similar" too. In this way, this score can be used to cluster a set of users into sets of "similar" users. It could be used in some recommendation algorithms or tune already existing algorithms to consider a cluster's peculiarities. Also, we can extend our model and calculate feature vectors for subreddits. In that way, we can find similar to the user's subreddits and recommend them to him.
{"title":"Practical OSINT investigation - Similarity calculation using Reddit user profile data","authors":"Valeria Vishnevskaya, Klaus Schwarz, Reiner Creutzburg","doi":"10.2352/ei.2023.35.3.mobmu-356","DOIUrl":"https://doi.org/10.2352/ei.2023.35.3.mobmu-356","url":null,"abstract":"This paper presents a practical Open Source Intelligence (OSINT) use case for user similarity measurements with the use of open profile data from the Reddit social network. This PoC work combines the open data from Reddit and the part of the state-of-the-art BERT model. Using the PRAW Python library, the project fetches comments and posts of users. Then these texts are converted into a feature vector - representation of all user posts and comments. The main idea here is to create a comparable user's pair similarity score based on their comments and posts. For example, if we fix one user and calculate scores of all mutual pairs with other users, we will produce a total order on the set of all mutual pairs with that user. This total order can be described as a degree of written similarity with this chosen user. A set of \"similar\" users for one particular user can be used to recommend to the user interesting for him people. The similarity score also has a \"transitive property\": if $user_1$ is \"similar\" to $user_2$ and $user_2$ is similar to $user_3$ then inner properties of our model guarantees that $user_1$ and $user_3$ are pretty \"similar\" too. In this way, this score can be used to cluster a set of users into sets of \"similar\" users. It could be used in some recommendation algorithms or tune already existing algorithms to consider a cluster's peculiarities. Also, we can extend our model and calculate feature vectors for subreddits. In that way, we can find similar to the user's subreddits and recommend them to him.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135694712","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.3.mobmu-368
Daniel Jaster, Reiner Creutzburg, Eberhard Hasche
In this work, the most relevant 3D LiDAR technologies and their applications in 2022 were investigated. For this purpose, applications of LiDAR systems were classified into the typical application areas "3D modeling", "smart city", "robotics", "smart automotive" and "consumer goods". The investigation has shown that neither "mechanical" LiDAR technologies, nor so-called solid-state LiDAR technologies, nor "hybrid" LiDAR technologies can be evaluated as optimal for the typical application areas. In none of the application areas could all of the elaborated requirements be met. However, the "hybrid" LiDAR technologies such as sequential MEMS LiDAR technology and sequential flash LiDAR technology proved to be among the most suitable for most typical application areas. However, other technologies also tended to be suitable for individual typical application areas. Finally, it was found that several of the LiDAR technologies investigated are currently equally suitable for some typical application areas. To evaluate the suitability, concrete LiDAR systems - of different technologies and properties - were compared with the specific requirements of exemplary applications of an application area. The results of the investigation provide an orientation as to which LiDAR technology is promising for which application area.
{"title":"A qualitative study of LiDAR technologies and their application areas","authors":"Daniel Jaster, Reiner Creutzburg, Eberhard Hasche","doi":"10.2352/ei.2023.35.3.mobmu-368","DOIUrl":"https://doi.org/10.2352/ei.2023.35.3.mobmu-368","url":null,"abstract":"In this work, the most relevant 3D LiDAR technologies and their applications in 2022 were investigated. For this purpose, applications of LiDAR systems were classified into the typical application areas \"3D modeling\", \"smart city\", \"robotics\", \"smart automotive\" and \"consumer goods\". The investigation has shown that neither \"mechanical\" LiDAR technologies, nor so-called solid-state LiDAR technologies, nor \"hybrid\" LiDAR technologies can be evaluated as optimal for the typical application areas. In none of the application areas could all of the elaborated requirements be met. However, the \"hybrid\" LiDAR technologies such as sequential MEMS LiDAR technology and sequential flash LiDAR technology proved to be among the most suitable for most typical application areas. However, other technologies also tended to be suitable for individual typical application areas. Finally, it was found that several of the LiDAR technologies investigated are currently equally suitable for some typical application areas. To evaluate the suitability, concrete LiDAR systems - of different technologies and properties - were compared with the specific requirements of exemplary applications of an application area. The results of the investigation provide an orientation as to which LiDAR technology is promising for which application area.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"208 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135694715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.9.ipas-a09
Abstract Image Processing: Algorithms and Systems continues the tradition of the past conference, Nonlinear Image Processing and Pattern Analysis, in exploring new image processing algorithms. Specifically, the conference aims at highlighting the importance of the interaction between transform-, model-, and learning-based approaches for creating effective algorithms and building modern imaging systems for new and emerging applications. It also reverberates the growing call for integration of the theoretical research on image processing algorithms with the more applied research on image processing systems.
{"title":"Image Processing: Algorithms and Systems XXI Conference Overview and Papers Program","authors":"","doi":"10.2352/ei.2023.35.9.ipas-a09","DOIUrl":"https://doi.org/10.2352/ei.2023.35.9.ipas-a09","url":null,"abstract":"Abstract Image Processing: Algorithms and Systems continues the tradition of the past conference, Nonlinear Image Processing and Pattern Analysis, in exploring new image processing algorithms. Specifically, the conference aims at highlighting the importance of the interaction between transform-, model-, and learning-based approaches for creating effective algorithms and building modern imaging systems for new and emerging applications. It also reverberates the growing call for integration of the theoretical research on image processing algorithms with the more applied research on image processing systems.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135695208","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.5.iriacv-a05
Abstract This conference brings together real-world practitioners and researchers in intelligent robots and computer vision to share recent applications and developments. Topics of interest include the integration of imaging sensors supporting hardware, computers, and algorithms for intelligent robots, manufacturing inspection, characterization, and/or control. The decreased cost of computational power and vision sensors has motivated the rapid proliferation of machine vision technology in a variety of industries, including aluminum, automotive, forest products, textiles, glass, steel, metal casting, aircraft, chemicals, food, fishing, agriculture, archaeological products, medical products, artistic products, etc. Other industries, such as semiconductor and electronics manufacturing, have been employing machine vision technology for several decades. Machine vision supporting handling robots is another main topic. With respect to intelligent robotics another approach is sensor fusion – combining multi-modal sensors in audio, location, image and video data for signal processing, machine learning and computer vision, and additionally other 3D capturing devices. There is a need for accurate, fast, and robust detection of objects and their position in space. Their surface, background, and illumination are uncontrolled, and in most cases the objects of interest are within a bulk of many others. For both new and existing industrial users of machine vision, there are numerous innovative methods to improve productivity, quality, and compliance with product standards. There are several broad problem areas that have received significant attention in recent years. For example, some industries are collecting enormous amounts of image data from product monitoring systems. New and efficient methods are required to extract insight and to perform process diagnostics based on this historical record. Regarding the physical scale of the measurements, microscopy techniques are nearing resolution limits in fields such as semiconductors, biology, and other nano-scale technologies. Techniques such as resolution enhancement, model-based methods, and statistical imaging may provide the means to extend these systems beyond current capabilities. Furthermore, obtaining real-time and robust measurements in-line or at-line in harsh industrial environments is a challenge for machine vision researchers, especially when the manufacturer cannot make significant changes to their facility or process.
{"title":"Intelligent Robotics and Industrial Applications using Computer Vision 2023 Conference Overview and Papers Program","authors":"","doi":"10.2352/ei.2023.35.5.iriacv-a05","DOIUrl":"https://doi.org/10.2352/ei.2023.35.5.iriacv-a05","url":null,"abstract":"Abstract This conference brings together real-world practitioners and researchers in intelligent robots and computer vision to share recent applications and developments. Topics of interest include the integration of imaging sensors supporting hardware, computers, and algorithms for intelligent robots, manufacturing inspection, characterization, and/or control. The decreased cost of computational power and vision sensors has motivated the rapid proliferation of machine vision technology in a variety of industries, including aluminum, automotive, forest products, textiles, glass, steel, metal casting, aircraft, chemicals, food, fishing, agriculture, archaeological products, medical products, artistic products, etc. Other industries, such as semiconductor and electronics manufacturing, have been employing machine vision technology for several decades. Machine vision supporting handling robots is another main topic. With respect to intelligent robotics another approach is sensor fusion – combining multi-modal sensors in audio, location, image and video data for signal processing, machine learning and computer vision, and additionally other 3D capturing devices. There is a need for accurate, fast, and robust detection of objects and their position in space. Their surface, background, and illumination are uncontrolled, and in most cases the objects of interest are within a bulk of many others. For both new and existing industrial users of machine vision, there are numerous innovative methods to improve productivity, quality, and compliance with product standards. There are several broad problem areas that have received significant attention in recent years. For example, some industries are collecting enormous amounts of image data from product monitoring systems. New and efficient methods are required to extract insight and to perform process diagnostics based on this historical record. Regarding the physical scale of the measurements, microscopy techniques are nearing resolution limits in fields such as semiconductors, biology, and other nano-scale technologies. Techniques such as resolution enhancement, model-based methods, and statistical imaging may provide the means to extend these systems beyond current capabilities. Furthermore, obtaining real-time and robust measurements in-line or at-line in harsh industrial environments is a challenge for machine vision researchers, especially when the manufacturer cannot make significant changes to their facility or process.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135695223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.12.ervr-217
Sharad Sharma, JeeWoong Park, Brendan Tran Morris
There is a need to prepare for emergencies such as active shooter events. Emergency response training drills and exercises are necessary to train for such events as we are unable to predict when emergencies do occur. There has been progress in understanding human behavior, unpredictability, human motion synthesis, crowd dynamics, and their relationships with active shooter events, but challenges remain. This paper presents an immersive security personnel training module for active shooter events in an indoor building. We have created an experimental platform for conducting active shooter drills for training that gives a fully immersive feel of the situation and allow one to perform virtual evacuation drills. The security personnel training module also incorporates four sub-modules namely 1) Situational assessment module, 2) Individual officer intervention module, 3) Team Response Module, and 4) Rescue Task Force module. We have developed an immersive virtual reality training module for active shooter events using an Oculus for course of action, visualization, and situational awareness for active shooter events as shown in Fig.1. The immersive security personnel training module aims to get information about the emergency situation inside the building. The dispatched officer will verify the active shooter situation in the building. The security personnel should find a safe zone in the building and secure the people in that area. The security personnel should also find the number and location of persons in possible jeopardy. Upon completion of the initial assessment, the first security personnel shall advise communications and request resources as deemed necessary. This will allow determining whether to take immediate action alone or with another officer or wait until additional resources are available. After successfully gathering the information, the personnel needs to update the info to their officer through a communication device.
{"title":"Immersive security personnel training module for active shooter events","authors":"Sharad Sharma, JeeWoong Park, Brendan Tran Morris","doi":"10.2352/ei.2023.35.12.ervr-217","DOIUrl":"https://doi.org/10.2352/ei.2023.35.12.ervr-217","url":null,"abstract":"There is a need to prepare for emergencies such as active shooter events. Emergency response training drills and exercises are necessary to train for such events as we are unable to predict when emergencies do occur. There has been progress in understanding human behavior, unpredictability, human motion synthesis, crowd dynamics, and their relationships with active shooter events, but challenges remain. This paper presents an immersive security personnel training module for active shooter events in an indoor building. We have created an experimental platform for conducting active shooter drills for training that gives a fully immersive feel of the situation and allow one to perform virtual evacuation drills. The security personnel training module also incorporates four sub-modules namely 1) Situational assessment module, 2) Individual officer intervention module, 3) Team Response Module, and 4) Rescue Task Force module. We have developed an immersive virtual reality training module for active shooter events using an Oculus for course of action, visualization, and situational awareness for active shooter events as shown in Fig.1. The immersive security personnel training module aims to get information about the emergency situation inside the building. The dispatched officer will verify the active shooter situation in the building. The security personnel should find a safe zone in the building and secure the people in that area. The security personnel should also find the number and location of persons in possible jeopardy. Upon completion of the initial assessment, the first security personnel shall advise communications and request resources as deemed necessary. This will allow determining whether to take immediate action alone or with another officer or wait until additional resources are available. After successfully gathering the information, the personnel needs to update the info to their officer through a communication device.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135694149","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.12.ervr-a12
Abstract Virtual and augmented reality systems are evolving. In addition to research, the trend toward content building continues and practitioners find that technologies and disciplines must be tailored and integrated for specific visualization and interactive applications. This conference serves as a forum where advances and practical advice toward both creative activity and scientific investigation are presented and discussed. Research results can be presented and applications can be demonstrated.
{"title":"Engineering Reality of Virtual Reality 2023 Conference Overview and Papers Program","authors":"","doi":"10.2352/ei.2023.35.12.ervr-a12","DOIUrl":"https://doi.org/10.2352/ei.2023.35.12.ervr-a12","url":null,"abstract":"Abstract Virtual and augmented reality systems are evolving. In addition to research, the trend toward content building continues and practitioners find that technologies and disciplines must be tailored and integrated for specific visualization and interactive applications. This conference serves as a forum where advances and practical advice toward both creative activity and scientific investigation are presented and discussed. Research results can be presented and applications can be demonstrated.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135695217","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-01-16DOI: 10.2352/ei.2023.35.3.mobmu-363
Kirsnaragavan Arudpiragasam, Taraka Rama Krishna Kanth Kannuri, Klaus Schwarz, Michael Hartmann, Reiner Creutzburg
Over the past decade, researchers have suggested many methods to find anomalies. However, none of the studies has applied frame reconstruction with Object Tracking (OT) to detect anomalies. Therefore, this study focuses on road accident detection using a combination of OT and U-Net associated with variants such as skip, skip residual and attention connections. The U-Net algorithm is developed for reconstructing the images using the UFC-Crime dataset. Furthermore, YOLOV4 and DeepSort are used for object detection and tracking within the frames. Finally, the Mahalanobis distance and the reconstruction error (RCE) are determined using a Kalman filter and the U-Net model.
{"title":"Improvement of vehicles accident detection using object tracking with U-Net","authors":"Kirsnaragavan Arudpiragasam, Taraka Rama Krishna Kanth Kannuri, Klaus Schwarz, Michael Hartmann, Reiner Creutzburg","doi":"10.2352/ei.2023.35.3.mobmu-363","DOIUrl":"https://doi.org/10.2352/ei.2023.35.3.mobmu-363","url":null,"abstract":"Over the past decade, researchers have suggested many methods to find anomalies. However, none of the studies has applied frame reconstruction with Object Tracking (OT) to detect anomalies. Therefore, this study focuses on road accident detection using a combination of OT and U-Net associated with variants such as skip, skip residual and attention connections. The U-Net algorithm is developed for reconstructing the images using the UFC-Crime dataset. Furthermore, YOLOV4 and DeepSort are used for object detection and tracking within the frames. Finally, the Mahalanobis distance and the reconstruction error (RCE) are determined using a Kalman filter and the U-Net model.","PeriodicalId":73514,"journal":{"name":"IS&T International Symposium on Electronic Imaging","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135695029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：