AccidentGPT: Large Multi-Modal Foundation Model for Traffic Accident Analysis

Traffic accident analysis is pivotal for enhancing public safety anddeveloping road regulations. Traditional approaches, although widely used, areoften constrained by manual analysis processes, subjective decisions, uni-modaloutputs, as well as privacy issues related to sensitive data. This paperintroduces the idea of AccidentGPT, a foundation model of traffic accidentanalysis, which incorporates multi-modal input data to automaticallyreconstruct the accident process video with dynamics details, and furthermoreprovide multi-task analysis with multi-modal outputs. The design of theAccidentGPT is empowered with a multi-modality prompt with feedback fortask-oriented adaptability, a hybrid training schema to leverage labelled andunlabelled data, and a edge-cloud split configuration for data privacy. Tofully realize the functionalities of this model, we proposes several researchopportunities. This paper serves as the stepping stone to fill the gaps intraditional approaches of traffic accident analysis and attract the researchcommunity attention for automatic, objective, and privacy-preserving trafficaccident analysis.