Risky Driving Behavior Identification Of Urban Road Sections Based On Hidden Markov Model

The urban road historical traffic accident statistics show that, the human factor occupies a dominant position in all the factors which cause traffic accidents. Based on historical accident data of urban road safety research itself is a remedial action after traffic accident, and the measure of based on non-accident data gradually arousing attention in the area of traffic safety. We can identify the urban road risky driving behavior from the common traffic flow information and traffic conflict effectively. It is an important way to evaluate urban road safety level of risky driving.This paper analyses the basic characteristics of urban road traffic flow, and dividing the traffic flow state into three categories, which is free flow state, steady flow state and unsteady flow state. Corresponding, there are three kinds of driving behavior, which is free driving behavior, following driving behavior and lane changing driving behavior. Furthermore, the risky driving behaviors of the urban road section are defined, which are speeding risky driving behaviors, line driving behaviors, illegal turn driving behaviors, did not keep safe distance away from the vehicle driving behaviors, frequent changing lane driving behaviors. Collect urban road section traffic flow video data of sunny days, rainy days respectively. Use the program software of vehicle tracking demo upgrade to obtain the motion of each vehicle trajectory data, and extract each vehicle’s motion parameters. Meanwhile, analysis the traffic flow characteristics and driving characteristics of research section of Chang Jiang Road. Then analysis traffic volume and its composition, vehicle speed, vehicle acceleration, time headway, conflict distance, the distance from the right side of the lane line and the distance from the camera of sunny days and rainy days. Conflict index is constructed in this paper, the conflict index is changing with the change of velocity, acceleration, weather and so on, which make up the shortcomings of the previous research on the assumption that the speed, direction, acceleration or deceleration is invariable.This paper use a left-to-right model structure of hidden Markov models, input observation sequence of velocity, acceleration, the distance from the right side of the lane line, the conflict distance and the distance from the camera five observation information. Then recognize the normal driving behavior, speeding risky driving behaviors, line driving behaviors, illegal turn driving behaviors, did not keep safe distance away from the vehicle driving behaviors, frequent changing lane driving behaviors six states. HMM risky driving behavior identification model were constructed on sunny days and rainy days. What’s more, verify its recognition accuracy is as high as 86%. Finally, this paper puts forward some measures to improve the performance of risky driving behavior, and the results can be evaluated by using HMM model.