Evaluation of Cooperative Perception Algorithms in Simulation Environments with Realistic Communication Models

Perception systems for automated vehicles frequently incorporate detection, tracking and segmentation tasks. A notable approach is cooperative perception (co-perception), utilizing vehicle-to-vehicle (V2V) communication to exchange individual perception data among multiple vehicles, enhancing the collective situational awareness on the road. the challenge faced by many co-perception models today is that they do not take into account realistic communication effects. Most co-perception models assume ideal communication or emulate possible data loss or delay. Since wireless communication data loss is inevitable due to the nature of radio signals, co-perception models must take factors such as signal loss or transmission delay into account. This article explores existing co-perception models that consider parameters related to realistic communication effects. the V2VNet, When2com, DiscoNet, and V2VAM models are considered. For each model, a description and analysis of their features is presented. Additionally, the use of recorded datasets for co-perception model verification and its implications for model validation is discussed. Based on the performed analysis, the use of a connected and automated vehicle integrated simulation environment is proposed for testing co-perception models with decision-making data feedback and realistic communications. the proposed simulation environment is verified for both co-perception model simulation and realistic communication data consideration.