基于人工智能建模的危险驾驶行为识别和风险评估

According to statistics of road traffic accident, driving behavior is very important for traffic safety. Dangerous driving behavior is the main cause of traffic insecurity. The key to the project is establishing a highly precious identification model and risk evaluation system for dangerous driving behaviors, which are of great significance for road safety decision-making. Based on the urban road network of Shanghai and the 13th Five Year National key research and development program, the driving behavior data of various motor vehicles under different road conditions and environment will be collected by several detection methods. The key characteristic parameters sets of dangerous driving behavior under different conditions will be established by combining the data sets of dangerous driving behavior and analysis of influence factors. Then the identification model for dangerous driving behaviors will be built by artificial intelligence algorithms including NN and SVM. The identification accuracy will be improved by the fusion of multi-sources data, which will broaden the research of dangerous driving behaviors. And the risk evaluation system will be built by quantifying the risk evaluation indicators to analyze the risk evolution trend of dangerous driving behaviors. The research results of this project will help to establish a dangerous driving risk management system with practical application significance. It is very important and meaningful for the improvement of active safety warning and management intervention level of road traffic.

根据道路交通历史事故统计数据显示，驾驶行为对于交通安全来说非常重要，危险驾驶行为是交通不安全的主要诱因。建立面向危险驾驶行为的高精度判别模型和风险评估体系是本项目的关键，对道路交通安全决策具有重要的意义。本研究依托上海城市道路网和十三五国家重点专项研发计划，通过各种检测手段收集不同道路条件和环境因素下的各类型机动车的驾驶行为数据；根据不同场景工况下的危险驾驶行为数据集，结合影响因素分析，构建不同场景工况下的危险驾驶行为关键特征参数集。在此基础上，应用神经网络和支持向量机等人工智能算法，建立面向危险驾驶行为的判别模型，通过驾驶数据进行多源融合，提高判别精度，拓宽危险驾驶行为判别的研究思路。量化风险评价指标，结合风险评估体系，解析危险驾驶行为的风险演变趋势。本项目的实施有助于建立有实际应用意义的危险驾驶风险管理体系，对道路交通的主动安全预警和管理干预技术水平的实质性提高是很重要和有意义的。

面对交通事故的频发和汽车保有量的激增，开展危险驾驶行为识别对降低驾驶风险和提高道路交通安全具有重要意义。本研究依托上海市、浙江省、和江苏省的道路交通网络，采用高空无人机拍摄和计算机视觉解析技术收集了大量驾驶行为数据。结合驾驶行为谱，构建了基于风险度量的危险驾驶行为量化方法，使用基于统计分布的四分位差法确定了危险驾驶行为的判断阈值。结合人工智能算法，以驾驶行为谱中的风险度量参数的关键特征值作为输入，构建了基于人工智能方法的危险驾驶行为辨识理论、框架和模型，实现实时标定道路行驶中的危险驾驶行为，为交通安全主动防控提供决策依据。通过驾驶行为风险源辨识，解析危险驾驶行为致灾机理，从而建立危险驾驶行为的风险评估指标体系，综合评估危险驾驶行为的事故概率和风险等级。研究成果将把握我国道路交通驾驶行为运行特征，明确危险驾驶行为的辨识框架与风险演变趋势，提出危险驾驶行为的防治措施和预警对策，提高我国道路交通主动安全防控的理论研究和决策应用的科学水平。