Research on Generalization Methods for Route Acquisition Data Based on Gaussian Mixture Models

With the development of intelligent driving technology, the safety verification of intelligent driving vehicles faces challenges. Traditional test and verification methods based on mileage have been unable to meet the requirements of verification of highly concurrent data of intelligent driving functions, and scenario-based verification theory has been gradually studied in the field of automatic driving. How to ensure the coverage and validity of scene data has become a hot research issue. In this paper, a generalization algorithm of road sampling data based on mixed Gaussian model is proposed. Based on the collected multi-type spatio-temporal trajectory data, the key parameters are fitted with mixed Gaussian distribution, and parameter correlation is analyzed to achieve standardization and normalization of action description. The mixed Gaussian model algorithm proposed in this paper has a fitting parameter accuracy of more than 95% and a generalization speed of 1,000 / min. After screening, it can form effective scenes corresponding to high confidence, and the confidence of dangerous scenes can reach 99%. In terms of technology, an automatic generation technology of full coverage simulation test cases is proposed to solve the pain point problem of insufficient coverage of simulation scenes. In terms of tool integration, the whole process solution from data to test cases to simulation is formed based on Open series standards.