Reconstruction of Accident Report of Autonomous Vehicle Based on Genetic Algorithm

The current autonomous driving acceleration test mainly relies on image or video data sets, which are difficult and costly to count and obtain. Therefore, from another perspective-using accident text data to accelerate autonomous driving testing, this paper proposes a combinatorial optimization based on genetic algorithms. Based on the accident report of self-driving cars in California, USA, the key scene information is defined and extracted for combinatorial optimization to obtain a series of new accident texts. Furthermore, the scene visualization of the generated new accident text is carried out, and the performance of the generated scene is evaluated. The experimental results show that compared with the original accident report, the average risk factor of the new accident report generated by the institute has increased by 32%, and the danger of the generated scene is much higher than that of the natural data set, which proves that the proposed in this paper is helpful Accelerate the construction of the autonomous driving scene test library from the source.