Aiming at the hassle of intent time window selection and intent characteristic parameters determination in driving intention recognition, two distinctive intention time window division strategies are proposed. The experiment was carried out in the driving simulator, and 160 units of valid sets of using samples were selected from the driving samples collected from 15 subjects, and the driving intentions were categorized into three categories: lane keeping (LK), lane changing left (LCL), and lane changing right (LCR). Pearson correlation analysis was performed on the intention characteristic parameters by comparing the differences in the intention samples and considering the correlation between the parameters. Thereafter six driving intention feature parameters were identified. Subsequently, the time of the vehicle's front wheel pressure point is calculated based on the yaw angle and the distance from the vehicle centroid to the lane centerline to determine the first intention time window. Simultaneously, the second intent window is determined by K-means clustering of the steering wheel angle and the distance from vehicle centroid to the lane centerline. Ultimately, the linear chain Conditional Random Field model is established to train and recognize the intent samples composed of six parameters and compare with HMM. The simulation results demonstrate that CRF has a good recognition effect, in which the second intention time window has the best recognition effect. The recognition accuracy of the three driving intentions of LCL, LK, and LCR reached 99.48%, 95.78%, and 98.74%, respectively. Specifically, the driver's driving intention can be accurately identified 1.35s before the start of a left lane change and 1.19s before the start of a right lane change. The study offers a treasured reference for the improvement of the area of driving intention recognition.