Abstract: The charge coupled device(CCD) installed in front of vehicle acquires road image in real-time, objective function is constituted by gray level and gradient character, and the road boundaries are fitted by using a parabola modal , so the identifications of the linear lane and the bent lane are unified. Identification is divided into three phases, the ROI of parabola parameters in each phase is set The parameters of parabola are optimized by using genetic algorithm. The parabola with approximately maximum objective function value is searched out through processes of initialization coding, calculating fitness, multi-point crossover and mutation, that is the optimal solution. The road identification result is used to obtain the direction deviation and lateral deviation of preview point and the road curvature, thus providing information for trajectory prediction. Based on the dynamic characteristics, the vehicle front steer angle of kinematics model is modified. The vehicle front steer angle and the road curvature are used as input of system, the preview dynamic continuous model is built. According to the sampling frequency of system, the continuous model is discretized. Kalman filter theory is used to design the state observer for observing real-time lateral speed and yaw velocity of vehicle, thus obtaining the vehicles trajectory. The tests show that the method not only simulates the boundaries of road exactly under complex environment and predicts vehicle trajectory accurately, but also has strong real-time performance.*

Key words: Genetic algorithm, Intelligent vehicle, Kalman filter, Road boundary identification, State observer, Vehicle trajectory prediction