UA cloudflare authentication

Browsing by Author "Jones, Steven L"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Application of Conventional and Deep-Learning Methods to Address Rollover Crashes in Namibia
    (University of Alabama Libraries, 2023) Bullard, Jeffrey Cailis; Jones, Steven L
    Road traffic crashes are a leading cause of serious injuries and fatalities globally. They also place unnecessary developmental and economic burdens on low- and middle-income countries (LMICs) as they account for most of the world's road related deaths. This is typically due to both the increased frequency of dangerous crash types and the increased severity of said crash types. Rollover crashes while quite rare are a particularly dangerous crash type among other various crash types. In the case of Namibia, rollover crashes reportedly accounted for 34% of both road related injuries and fatalities in Namibia for 2020. Therefore, it crucial to understand the contributing factors and their associated effects on rollover crash severities in these countries. This thesis aims to investigate the significant factors influencing crash severities and their associated impact magnitudes on single vehicle rollover crashes in Namibia by adopting a mixed logit with heterogeneity in means and variances approach for 2014-2016 historical crash records. Additionally, roadside safety features such as paved road shoulder are a crucial component in the mitigation of these rollover crash types. While the lack of road safety features is an issue in LMICs, so is the ability to map the presence of said features. This a case study on the B2 highway in Namibia, aiming to classify images of the road according to the size of the road shoulder is also conducted. Google Street View images are labeled based to their road shoulder quality (none, up to two feet, larger than two feet). Based on the labeled images, we train a deep-artificial-neural-network to classify images according to those labels. Results indicate that rollover crashes in Namibia are influenced by several explanatory variables. Additionally, results from the deep-learning model have the potential to lower costs of mapping road safety features in LMIC.