Abstract
In this research, a novel technique for virtual lane detection to help enhance road safety is introduced. A virtual lane is a partition of the road that one or more consecutive vehicles should occupy on the road and be considered following the same path within a hypothetical right and left lane borders. The virtual lanes detected are called “practical-lanes”. The technique is developed for unstructured roads (roads with no lane markings) and for partially occluded roads. Prior to the development, a survey of the available and relevant literature is made to identify the achievements and challenges in the field.
The research hypothesis is that practical-lane detection is essential for enhancing road safety, as it will address a set of challenges mentioned in literature but not sufficiently researched, such as lane detection in unstructured roads and medium-to-high-density traffic roads. In addition, introducing a practical-lane concept and detection will enhance road safety by enabling the research question: What are the virtual lanes that cars are assuming on the road? The research objectives are to 1. Develop a framework for a new research topic: The Practical-Lane Detection. 2. Develop an algorithm to model practical-lanes based on vehicles observed on the road with respect to a reference car. 3. Implement a solution on a workstation capable of processing video feed in real-time.
The designed solution configuration is adapted from the generic configuration for lane detection introduced in the literature.
Data collection was by employing a camera phone, 26 video captures were taken in unstructured and partially occluded road scenes with varying lighting conditions. Data collection was designed to be representative of the intended road scenes by taking the captures in diverse districts of the city of Cairo and its surrounding highways. No specific challenges were faced during data collection. Validation and verification procedures were designed to test the developed solution on qualitative and quantitative dimensions.
An analysis of the results and system performance is made to conclude the key results and findings. The performance metrics used to evaluate the system’s performance are accuracy, precision, recall, and FPS. Key results for system performance are average accuracy of 73%, 100% precision, and 73% recall for linear roads. The average processing speed achieved is 30 FPS versus the real-time requirement of 30 FPS processing speed, so the real-time requirement is met by the solution. Statistical methods used to validate the findings were the mean and standard deviation for the performance metrics results, and the correlation coefficient for observed trends.
The concept introduced (practical-lanes) and the solution developed (practical-lanes detector) allow computer systems to detect driving lanes based on vehicle positions (virtual lanes). The potential real-world applications are 1. an expanded scope of road scenarios (unstructured & partially occluded), and 2. the ability to detect the actual lanes that vehicles are assuming independent of the lane marking on the road. In systems such as lane departure warning systems, lane departure assistance systems, and autonomous car projects, this contribution will impact road safety improvement. Key findings are that the system developed can detect virtual lanes on unstructured and partially occluded roads which has not been considered in literature yet.
The developed solution introduces a concept – practical-lanes – to the lane detection field that has not been considered in the literature. The existing lane detection methods use lane markings to detect driving lanes. Therefore, the contribution is an improvement to the existing solutions because the solution provides a dynamic picture of virtual lanes on the road versus a picture of the painted lane markings which might not be followed by vehicles on the road.
The research addresses the limitations of the lane detection field by operating on unstructured and partially occluded roads which is not addressed sufficiently in the field. The limitation of the developed solution is that it uses a linear model for practical-lane modelling, which makes the system outcomes better suited for straight and slightly curved roads. Also, the KPIs’ calculation was done manually which might affect reliability of calculations. Potential improvements to the developed model are to track the detected practical-lanes across frames, consideration of vehicle trajectory and orientation, and develop a non-linear model for practical-lanes to enhance the model’s performance in curved roads. Also, the automation of KPIs’ calculation will improve results reliability.
School
School of Sciences and Engineering
Department
Mechanical Engineering Department
Degree Name
MS in Mechanical Engineering
Graduation Date
Spring 6-15-2024
Submission Date
2-19-2024
First Advisor
Maki K. Habib
Committee Member 1
Amr Goneid
Committee Member 2
Heba El-Nemr
Extent
185p.
Document Type
Master's Thesis
Institutional Review Board (IRB) Approval
Not necessary for this item
Recommended Citation
APA Citation
Shehata, B. M.
(2024).Bola Shehata_Graduate Thesis_Practical-Lane Detection for Medium-to-High Density Traffic in Unstructured or Partially Occluded Roads [Master's Thesis, the American University in Cairo]. AUC Knowledge Fountain.
https://fount.aucegypt.edu/etds/2309
MLA Citation
Shehata, Bola Marei Kamel M. Sc.. Bola Shehata_Graduate Thesis_Practical-Lane Detection for Medium-to-High Density Traffic in Unstructured or Partially Occluded Roads. 2024. American University in Cairo, Master's Thesis. AUC Knowledge Fountain.
https://fount.aucegypt.edu/etds/2309