Infrastructure Assisted Cooperative and Distributed Perception Strategies for Connected Vehicles

The research presented in this dissertation addressed the challenge of efficiently processing sensor data across vehicles, edge, and cloud platforms to support resource-intensive perception tasks in autonomous driving. We investigated distributed processing of perception data from three perspectives to enhance perception accuracy, optimize network and computing resources, and meet realtime
latency constraints: (1) offloading computationally intensive tasks to edge or cloud
platforms for greater processing capacity and reduced on-board load, (2) evaluating the
impact of compression techniques such as H.265 and JPEG on perception quality and
transmission latency, and (3) exploring feature-vector transmission as an alternative to
raw or compressed data to reduce bandwidth usage while maintaining perception accuracy and minimizing end-to-end delay. To evaluate performance under realistic conditions, the study utilized both simulationbased assessments and real-world experiments. Using Vehicle-to-Everything (V2X) communication technologies, we analyzed compression and feature-sharing strategies for seamless data exchange between vehicles and infrastructure. Our findings demonstrated that the proposed cooperative and distributed perception strategies significantly improved detection
accuracy and reduced processing delays compared to standalone on-board systems.