An Edge Intelligence-Empowered Metaverse Framework for Intelligent Transportation Systems

Date of Award


Document Type


Degree Name

Master of Science in Machine Learning


Machine Learning

First Advisor

Dr. Mohsen Guizani

Second Advisor

Dr. Martin Takac


With the advancement in modern computational technology, AI-based automation in transportation is gaining an unprecedented popularity among common lifestyle. This trend has forced the development of novel vehicular network applications, for example, lane change assistance, collision avoidance, accident reporting, and infotainment. Safe and reliable operations of autonomous vehicles with such design poses many challenges for the researchers. It poses the need of further investigation on the issues of resource management and configurations of metaverses at the network edge, considering factors such as traffic density, coverage, latency, and network capacity to ensure seamless communication and efficient operation for vehicular applications. Also, for automatic operations of such vehicular applications, there is need for automatic recognition of traffic signs irrespective of language of the text on traffic sign boards for enhancing the autonomy and safety of intelligent transportation systems. This thesis presents an edge intelligence-enabled metaverse with deep learning framework for automatic traffic sign recognition for safety and reliable operations. Enabling the metaverse for a vehicular network requires careful placement of meta spaces (i.e., virtual machines running a virtual model of the actual network) at the network edge. Therefore, a framework for the efficient placement of distributed meta spaces for various applications using the concept of virtual machines. For this, We formulate a cost function that accounts for latency (i.e., meta space computing latency and transmission latency for meta spaces signaling). To minimize this cost, an optimization problem is formulated that uses three variables: (a) meta spaces (i.e., based on virtual machines) operating frequency, (b) meta space placement, and (c) wireless resource allocation. A decomposition-based solution is used to solve the formulated problem due to its difficult nature. For automatic recognition and interpretation of road traffic signs, a novel vision transformer architecture has been proposed that uses the scale-invariant feature transforms (SIFT) to create the image patches and sequence embedding. The proposed framework combines the feature extraction capabilities of the SIFT algorithm with the capability of Vision Transformer (ViT) to capture the sequential dependencies of the obtained features. The ViT-based framework capable of detecting and interpreting the traffic signs in diverse environmental conditions, thereby enhancing the autonomy and safety of intelligent transportation systems. The proposed method is validated on the German Traffic Sign Recognition Benchmark (GTSRB) and UAE Traffic Sign Recognition dataset (UTSRD). The validation results show that the technique can achieve up to classification accuracy of 98.5% and 98.2% for the GTSRB and UTSRD, respectively.


Thesis submitted to the Deanship of Graduate and Postdoctoral Studies

In partial fulfilment of the requirements for the M.Sc degree in Machine Learning

Advisors: Mohsen Guizani, Dr. Martin Takac

Online access available for MBZUAI patrons