Research on Prediction Algorithm of Vehicle Trajectory in Front Based on Driving Intention Classification

Abstract

Predicting the trajectory of surrounding vehicles has an essential impact on the vehicle's future decision-making and path planning and helps to improve the driving safety, fuel consumption, and traffic efficiency of the vehicle. However, due to the uncertainty of its driving intention, the interaction between the predicted object and the surrounding environment, its trajectory prediction faces enormous challenges. In this paper, based on the Fuzzy C-means algorithm, the automatic recognition of driving intention is realized through offline training by using the information about the rate of change of the heading angle. Combined with vehicle speed information, a trajectory prediction algorithm based on Long short-term memory is developed, and the results of prediction models trained for different driving intentions are merged to obtain the future trajectory of the vehicle. The artificial potential field method is used to calculate the target vehicle's longitudinal and lateral safety distance to summarize the potential field of the road boundary and the surrounding vehicles of the predicted object to achieve the effect of correcting the trajectory prediction result. The 1s rolling prediction is carried out by the iterative method. Among the horizontal distances, 94.52% of the prediction errors are less than 0.1m, and the maximum distance prediction error is 0.23m. In the longitudinal distance, 92.91% of the prediction errors are less than 0.5m, and the maximum distance prediction error is 0.87m.