Real-time comprehensive driving ability evaluation algorithm for intelligent assisted driving

Abstract

To meet the needs of the human-machine co-driving decision problem in the intelligent assisted driving system for real-time comprehensive driving ability evaluation of drivers, this paper proposes a real-time comprehensive driving ability evaluation method that integrates driving skill, driving state, and driving style. Firstly, by analyzing the driving experiment data obtained based on the intelligent driving simulation platform (the experiment can effectively distinguish the driver's driving skills and avoid the interference of driving style), the feature values that significantly represent driving skills and driving state are selected, and the time correlation between driving state and driving skills is pointed out. Furthermore, the concept of relativity in comprehensive driving ability evaluation is further proposed. Under this concept, the natural driving trajectory dataset-HighD is used to establish the distribution map of feature values of the human driver group as the evaluation benchmark to realize the relative evaluation of driving skill and driving state. Similarly, HighD is used to establish a distribution map of human driver style feature values as an evaluation benchmark to achieve relative driving style evaluation. Finally, a comprehensive driving ability evaluation model with a “punishment” and “affirmation” mechanism is proposed. The experimental comparative analysis shows that the evaluation algorithm proposed in this paper can take into account the driver's driving skill, driving state, and driving style in the real-time comprehensive driving ability evaluation, and draw differential evaluation conclusions based on the “punishment” and “affirmation” mechanism model to achieve a comprehensive and objective evaluation of the driver's driving ability. It can meet the needs of human-machine shared driving decisions for driver's driving ability evaluation.

Keywords

• Driving ability evaluation
• Driving style
• Unsupervised decision tree
• Gaussian kernel function
• Human-machine shared control