Research On Key Technology Of Driving Posture Understanding System

With automatic driving technology carrying on step by step,many problems has been exposed.Self-driving cars are armed with expensive sensors just to be able to make information redundant and thus to improve driving safety.However,it is not enough to improve driving safety in terms of sensors.Road conditions can change rapidly and dangers can occur in short time,so a long term consideration is needed.In order to better grasp the road conditions and provide more favorable information for making decisions,people have studied all kinds of traffic scenes and opened up the research direction of driving situation understanding,such as dynamic trajectory prediction and driving intention understanding.The premise of dynamic trajectory prediction is to detect and track the target.The Kalman filter algorithm and the target association algorithm are used to estimate the latest trajectory,and a reasonable future trajectory is obtained by combining the characteristics of the trajectory curve and the probability model.There are various methods of driving intention prediction,the purpose of which is to estimate the future intention of the vehicle.Since intention cannot be observed directly,it is better to use Hidden Markov Model to predict it indirectly.On the road surface with many vehicles,it is necessary to give early warning to the dangerous situation in time to achieve the purpose of improving safety.Based on the robot research group of Jilin university,this paper studies the key technologies needed to build the driving situation understanding system.From actual driving scene,combined with the research of the existing research results and experimental platform,this paper proposes a method for constructing driving situation understanding system,realizes the dynamic trajectory prediction and driving intention prediction of the surrounding vehicles in the traffic scene,and aiming at the condition of the predicted results and their own vehicles,gives warning of driving posture in dangerous situations.The innovation of this paper is that compared with the traditional dynamic trajectory prediction algorithm,the curve fitting method is introduced to calculate the curvature of the vehicle,and the reasonable prediction of the trajectory sequence is completed.Through the experiments on different traffic scenes and data sets of different scales,it is verified that the dynamic trajectory prediction algorithm and driving intention understanding algorithm proposed in this paper can complete the construction of driving situation understanding system,and can give early warning in case of danger.