Intelligent Vehicle's Motion Perception And Lane Change Intention Recognition Of Vehicles Ahead

Intelligent vehicles is a vehicle that automatically senses the surrounding environment and automatically navigates without human intervention.Smart cars have many advantages such as improving driving safety,reducing traffic accidents,effectively managing traffic flow,and mitigating traffic pressure.Now they have become the main development direction of future automotive technology.The premise of intelligent vehicles driving safely on the road is to have an accurate perception of the driving environment,which is especially important for the understanding of the motion perception and driving behavior of other vehicles in transit.Based on the urban road environment,this paper studies the motion state and the intention of changing the vehicle target in front of the intelligent vehicle.The main research contents are as follows:Firstly,the road segment semantic segmentation technology based on deep convolutional neural network is studied.A lot of traffic road scene samples are collected and marked.The constructed deep road scene segmentation network is used to identify all the pixel categories in the road scene image through sample training,complete the smooth division of roads,vehicles,pedestrians,sidewalks,various lane lines,sky and other categories.Based on the results of semantic segmentation,the image RGB segmentation technology is used to extract the vehicles in the scene,and the rough positioning of the vehicle target and the preliminary extraction of the longitudinal/lateral velocity and acceleration motion parameters are completed based on the monocular vision ranging technology.Secondly,based on the relative motion relationship between the world coordinate system and the vehicle's motion coordinate system,it is considered that the maneuvering behavior of the road-driving vehicle is mainly based on the plane two-dimensional motion,and the mobility is small,and the moving target state is adopted.The comparison analysis between the model and the filter estimation algorithm finally uses the extended Kalman filter algorithm based on uniform acceleration model to realize the accurate real-time estimation of the position and longitudinal/lateral velocity and longitudinal/lateral acceleration of the target vehicle based on the world coordinate system.The stability and estimation accuracy of the algorithm are verified by real vehicle experiments.Finally,based on the principle that the driving decision of the vehicle conforms to the Markov process and the invisibility and randomness of the lane change behavior of the vehicle,a mixed Gaussian hidden Markov model(GM-HMM)is established to describe the lane change behavior of the preceding vehicle..Combined with the driving characteristics of the vehicle on the straight road,the lateral distance between the preceding vehicle and the left lane line measured by the sensor and the transverse longitudinal speed of the preceding vehicle are used as the parameters for the lane change behavior.The NGSIM data set is selected as the training sample to establish the preceding vehicle lane change.The intention identification model identifies the driving intentions of lane keeping(LK),left lane changing(LCL)and right lane changing(LCR)of the straight line segment.The recognition accuracy of the model is verified by testing the selected test samples.