Identifying drunk driving behavior on urban typical road section

The random, contacting and non-real-time test method of alcohol for drivers cannot satisfy the current situations now. Under the typical traffic conditions of urban road during accelerating, speed-keeping and turning section, vehicle speed, acceleration, position of gas pedal, and revolving speed of engine were taken as input variables. And a Support Vector Machine (SVM) model was introduced to identify the drivers' behaviors and to estimate whether the driver was under alcohol influence. To improve the training efficiency of the model, Particle Swarm Optimization (PSO) algorithm was used to optimize the parameters selection progress. The result proves that the SVM model optimized by PSO can quickly estimate whether the driver is under alcohol influence with a high accuracy. The model provides a theoretic support for the non-contact alcohol testing, leading to a practical application in the safety assistant driving system.