Date of Award

2-28-2024

Publication Type

Thesis

Degree Name

M.A.Sc.

Keywords

Deep Learning;Intelligent Transportation Systems;Lane Change Prediction;Privacy-preserving;Recurrent Neural Network;Secure Multiparty Computation

Supervisor

Majid Ahmadi

Supervisor

Shahpour Alirezaee

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract

Intelligent Transportation Systems (ITS) play a critical role in modern mobility, leveraging Advanced Driver Assistance Systems (ADAS) such as Lane Change Prediction (LCP) to improve navigation. Deep Learning Models have significantly enhanced LCP accuracy, but the need for extensive data poses privacy challenges in deploying such models. This thesis addresses these concerns by implementing a sequential-based neural network and Secure Multiparty Computation (SMPC) to balance privacy and performance. SMPC ensures no performance decline compared to other privacy methods like homomorphic encryption or differential privacy. The research utilizes the HighD dataset, known for its accuracy in autonomous driving applications, to deploy an RNN model for LCP. Performance evaluations are conducted before and after the application of privacy-preserving techniques. Additionally, the study examines how processing power affects model performance using Jetson devices, revealing that weaker devices can act as system bottlenecks. The findings underscore the importance of privacy-preserving methods in real-time applications like autonomous driving, highlighting the practicality and reliability of the proposed model in maintaining both privacy and performance standards. Through the utilization of SMPC, the study ensures robust privacy protection while maintaining model efficiency, contributing to the advancement of secure and efficient Deep Learning Models in intelligent transportation systems.

Download

Share

COinS