The evolution of detection systems and their application for intelligent transportation systems: From solo to symphony

Abstract

The emergence of autonomous driving technologies has been significantly influenced by advancements in perception systems. Traditional single-agent detection models, while effective in certain scenarios, exhibit limitations in complex environments, necessitating the shift towards collaborative detection models. While numerous studies have investigated the fundamental architecture and primary elements within this domain, comprehensive analyses focusing on the evolution from single-agent-based detection systems to collaborative detection systems are notably absent. This paper provides a comprehensive examination of this transition, delineating the development from single agent to collaborative perception models in autonomous driving. Initially, this paper delves into single-agent detection models, discussing their capabilities, limitations, and application scenarios. Subsequently, the focus shifts to collaborative detection models, which leverage Vehicle-to-Everything (V2X) communication to enhance perception and decision-making in complex environments. Fundamental concepts about mainstream collaborative approaches and mechanisms are reviewed to present the general organization of collaborative detection models. Furthermore, we critically evaluates various collaborative models, comparing their performance, data fusion strategies, and adaptability in dynamic settings. The integration of V2X-enabled Internet-of-Vehicles (IoV) introduces a pivotal evolution in the transition from single-agent-based detection to multi-agent collaborative sensing. This advancement allows for real-time interaction of sensory information between vehicles, augmenting the development of collaborative sensing. However, the interaction of sensory information also increases the load on the network, highlighting the need for strategies that achieve a balance between communication overhead and the improvement in perception capabilities. We concludes with future perspectives, emphasizing the potential issues the development of collaborative detection models will meet and the promising directions for future research.