An IVIS Typical Scene Generation Algorithm Based on Traffic Big Data

With the increasing development of automatic driving technology, the construction of virtual simulation scenes libraries for automatic driving vehicles, as well as the optimization of coverage of functions and redundancies based on the current scenes libraries, has become a problem that needs to be solved in process of the establishing the Intelligent Vehicle-Infrastructure System (IVIS) test and evaluation system, especially facing the uncountable inexhaustible library of actual traffic scenes. The construction of a standardized general test scene library covering typical scene applications, to provide a complete closed loop for automated driving vehicle testing also becomes a necessity. Based on the traffic big data, this paper takes the intelligent vehicle road system IVIS as the background, and aims to use scene essential factors, which are indecomposable factors obtained by scene decomposition, to describe the traffic scenes data, with feature extraction algorithm notion of unsupervised learning and nonlinear dimensionality reduction as a reference. Choosing traffic cell modelling, this paper adopts the idea of essential factors as the core and raises a vectorization process of scenes data as a foundation of subsequent research of the generation algorithms. On the basis of primitive scene decomposition, two core goals are set: study the typical test scenes of IVIS High-fidelity and flexible reconstruction technology and research on the scalable and easy-to-test generation technology of IVIS extreme test scenes. With scene vectorization process as the fundament, this paper attempts to select and optimize a suitable clustering algorithm, to use an improved density clustering algorithm called OIR-DBSCAN, to generate IVIS typical scenes, and accordingly ensure the generalizability and timeliness of the IVIS test.