Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Measurement of Bicycle–Vehicle Adjacency Distance Using Deep Neural Network
https://orcid.org/http://orcid.org/0009-0006-1474-108X
https://orcid.org/http://orcid.org/0000-0003-4318-421X
The lateral adjacent distance (LAD) between a bicycle and a passing vehicle is an important parameter that reflects the level of safety and comfort experienced by the cyclist. This LAD can be influenced by a variety of geometric design decisions. This study presents an automated technique to measure a LAD using deep learning (DL). Specifically, this paper studies the use of deep neural networks (DNN) to automatically measure bicycle–vehicle lateral distance as they pass each other on urban streets in the City of Ottawa. The sensor used is a camera, set-up at a relatively low altitude. Instance Segmentation (IS) is a technique that aims to accurately detect the outlines of objects of interest with varying appearances. IS has been successfully used in applications beyond road safety analysis, e.g., autonomous vehicles and medical scan diagnosis. In this study, 14 classes of road users are defined based on their functional attributes rather than their unique appearances. After the annotation of training examples for each class, the data is used to train a DNN till an acceptable learning rate is achieved. Subsequently, a lateral distance is measured between polygons of interest of the detected IS output object of interest and showed promising results. The developed model is an IS multi-class multi-object detector and classifier. The trained DNN is validated based on manually measured distances from multi-day video observation. Low-altitude cameras provide a challenge due to a higher level of occlusion compared to higher-altitude cameras. However, IS demonstrated the potential to overcome this challenge. A limitation of this study is that it does not automatically detect adjacency but, instead, measures the distance between a vehicle and a bicycle when the adjacency is observed. This model can help researchers in analyzing objective as well as subjective cyclist safety in urban areas.
Measurement of Bicycle–Vehicle Adjacency Distance Using Deep Neural Network
https://orcid.org/http://orcid.org/0009-0006-1474-108X
https://orcid.org/http://orcid.org/0000-0003-4318-421X
The lateral adjacent distance (LAD) between a bicycle and a passing vehicle is an important parameter that reflects the level of safety and comfort experienced by the cyclist. This LAD can be influenced by a variety of geometric design decisions. This study presents an automated technique to measure a LAD using deep learning (DL). Specifically, this paper studies the use of deep neural networks (DNN) to automatically measure bicycle–vehicle lateral distance as they pass each other on urban streets in the City of Ottawa. The sensor used is a camera, set-up at a relatively low altitude. Instance Segmentation (IS) is a technique that aims to accurately detect the outlines of objects of interest with varying appearances. IS has been successfully used in applications beyond road safety analysis, e.g., autonomous vehicles and medical scan diagnosis. In this study, 14 classes of road users are defined based on their functional attributes rather than their unique appearances. After the annotation of training examples for each class, the data is used to train a DNN till an acceptable learning rate is achieved. Subsequently, a lateral distance is measured between polygons of interest of the detected IS output object of interest and showed promising results. The developed model is an IS multi-class multi-object detector and classifier. The trained DNN is validated based on manually measured distances from multi-day video observation. Low-altitude cameras provide a challenge due to a higher level of occlusion compared to higher-altitude cameras. However, IS demonstrated the potential to overcome this challenge. A limitation of this study is that it does not automatically detect adjacency but, instead, measures the distance between a vehicle and a bicycle when the adjacency is observed. This model can help researchers in analyzing objective as well as subjective cyclist safety in urban areas.
Measurement of Bicycle–Vehicle Adjacency Distance Using Deep Neural Network
https://orcid.org/http://orcid.org/0009-0006-1474-108X
https://orcid.org/http://orcid.org/0000-0003-4318-421X
The lateral adjacent distance (LAD) between a bicycle and a passing vehicle is an important parameter that reflects the level of safety and comfort experienced by the cyclist. This LAD can be influenced by a variety of geometric design decisions. This study presents an automated technique to measure a LAD using deep learning (DL). Specifically, this paper studies the use of deep neural networks (DNN) to automatically measure bicycle–vehicle lateral distance as they pass each other on urban streets in the City of Ottawa. The sensor used is a camera, set-up at a relatively low altitude. Instance Segmentation (IS) is a technique that aims to accurately detect the outlines of objects of interest with varying appearances. IS has been successfully used in applications beyond road safety analysis, e.g., autonomous vehicles and medical scan diagnosis. In this study, 14 classes of road users are defined based on their functional attributes rather than their unique appearances. After the annotation of training examples for each class, the data is used to train a DNN till an acceptable learning rate is achieved. Subsequently, a lateral distance is measured between polygons of interest of the detected IS output object of interest and showed promising results. The developed model is an IS multi-class multi-object detector and classifier. The trained DNN is validated based on manually measured distances from multi-day video observation. Low-altitude cameras provide a challenge due to a higher level of occlusion compared to higher-altitude cameras. However, IS demonstrated the potential to overcome this challenge. A limitation of this study is that it does not automatically detect adjacency but, instead, measures the distance between a vehicle and a bicycle when the adjacency is observed. This model can help researchers in analyzing objective as well as subjective cyclist safety in urban areas.
Measurement of Bicycle–Vehicle Adjacency Distance Using Deep Neural Network
Lecture Notes in Civil Engineering
Desjardins, Serge (Herausgeber:in) / Poitras, Gérard J. (Herausgeber:in) / Siyoufi, Houssam (Autor:in) / Ismail, Karim (Autor:in)
Canadian Society of Civil Engineering Annual Conference ; 2023 ; Moncton, NB, Canada
Proceedings of the Canadian Society for Civil Engineering Annual Conference 2023, Volume 2 ; Kapitel: 11 ; 147-161
20.08.2024
15 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Vehicle Routing Problem Using Genetic Algorithms Based on Adjacency Relations
| British Library Conference Proceedings | 1995
Target Adjacency Effect Estimation Using Ground Spectrum Measurement and Landsat-5 Satellite Data
| Online Contents | 2006
Generation of rectangular dissections using directed adjacency graph
| British Library Online Contents | 2001