| dc.description.abstract | This dissertation introduces novel approaches to develop a comprehensive model to address situation awareness in the Internet of Cars, called Attention Assist Framework (AAF). The proposed framework utilizes both Low-Level Data Fusion (LLDF), and High-Level Information Fusion (HLIF) to implement traffic entity, situation, and impact assessment, as well as decision making.
The Internet of Cars is the convergence of the Internet of Things and Vehicular Ad-hoc Networks (VANETs). In fact, VANETs are the communication platforms that make possible the implementation of the Internet of Cars, and has become an integral part of this research field due to its major role to improve vehicle and road safety, traffic efficiency, and convenience as well as comfort to both drivers and passengers. Significant amount of VANETs research work has been focused on specific areas such as safety, routing, broadcasting, Quality of Service (QoS), and security. Among them, road safety issues are deemed one of the most challenging problems of VANETs. Specifically, lack of proper situational awareness of drivers has been shown to be the main cause of road accidents which makes it a major factor in road safety.
The traffic entity assessment relies on a LLDF framework that is able to incorporate various multi-sensor data fusion approaches with means of communication links in VANETs. This is used to implement a cooperative localization approach through fusing common data fusion methods, such as Extended Kalman Filter (EKF) and Unscented Transform (UT), and vehicle-to-vehicle communication in VANETs. Furthermore, traffic situation assessment is based on a fuzzy extension to the Multi-Entity Bayesian Networks (MEBNs), which exploit the expressiveness of first-order logic for semantic relations, and the strength of the Fuzzy Bayesian Networks in handling uncertainty, while tackling the inherent vagueness in the soft data created by human entities. Finally, the impact assessment and decision making is realized through incorporating notions of game theory into Fuzzy-MEBNs, and introducing Active Fuzzy-MEBN (ATFY-MEBN), which is capable in hypothesizing future situations by assessing the impact of the current situation upon taking the actions indicated by an optimal strategy. In fact, such strategies are achieved through solving the games that are generated through a novel situation-specific normal form games generation algorithm that aims to create games based on the given context. In general, ATFY-MEBN presents the concepts of players and actions, and includes new game components, along with a 2-tier architecture, to efficiently model impact assessment and decision making.
To demonstrate the capabilities of the proposed framework, a collision warning system simulator is
developed, which evaluates the likelihood of a vehicle being in a near-collision situation using a wide
variety of both local and global information sources available in the VANETs environment, and suggests an optimal action by assessing the impact of the current situation through generating and solving situation-specific games.
Accordingly, first, the entities that highly influence the safety aspect, as well as both their casual and semantic relationships are identified. Next, an ATFY-MEBN-based model is presented, which allows for modeling these entities along with their relationships in specific contexts, assessing the current states of the situations of interest, predicting their future states, and finally suggesting optimal decision.
Therefore, if the likelihood of being in a near-collision situation is determined to be high, and if the relevant situation-specific game is generated, then the impact of deciding on different combinations of actions that the game players take are calculated through a pre-fixed payoff function. Finally, the completed game is solved by finding its dominant strategy, that subsequently, results in proposing the optimal action to the driver.
Our experimental results are divided into three main sections, through which we evaluate the capabilities of the traffic entity, situation, and impact assessment methods. Accordingly, the performance of the proposed cooperative localization approach is assessed by comparing its results with the ground truth solution and that of the other localization methods in various driving test cases. Moreover, two distinct single-vehicle and multi-vehicles categories of driving scenarios, as well as a novel hybrid MEBN inference, demonstrate the capabilities of the proposed traffic assessment model to efficiently achieve situation and threat assessment on the road. Finally, the impact assessment and decision making models are evaluated through two different scenarios of driving in highway and intersection that are formed with various number of player vehicles, and their actions. | en |
