Research on Online Identification Algorithm of Dangerous Driving Behavior

Aiming at high false positives rate of dangerous driving behavior online identification system and high cost of data acquisition instruments at present, an algorithm to identify dangerous driving behavior by using variance Bayesian network was proposed. Experts put forward using variance Bayesian network model to identify dangerous behavior .Making use of on-board CCD to collect real-time road digital image and to determine the space to lane line of the front wheel according to vehicle parameters and vehicle space before and after combined with the probe installation angle and height. To set a model for particular lane deviation and with close car-following behavior according to vehicle space and lane line distance, and to determine the best time window length according to model recognition under each time window. Use Kalman filter to fit data in single window, to make sure the model output by determine the variance changes ,variance model output as the Bayesian model input, combine the result of variance model recognition with predicted result of Bayesian network, finally conduct the conclusion whether there exists danger in current driving behavior. At last, to have a test on model by using the real car test data, the best result indicates that existing model can identify two kings of dangerous driving behavior, besides, has better generalization than single variance model.