Multi Source Sensor Information Fusion And Analysis For Active Lane Changing

In order to realize the vehicle active lane change,we need to sense the surrounding environment and give the sensing results.Under this background,this paper studies how to realize the multi-source sensor information fusion and analysis.Based on a mass production vehicle project,this paper studies the target level fusion scheme and motion analysis algorithm under the sensor layout of four angle radars,one forward radar and one forward camera.Multi sensor fusion algorithm includes sensor data preprocessing module,multi-target fusion module and multi-target tracking module.The sensor data preprocessing module completes the space synchronization of multisensor targets;the Kalman filter model is established to solve the time synchronization of multi-sensor targets;the information of the forward camera is extended reversely and the curvature smoothing filter is used to reconstruct the backward lane line.In the multi-target fusion module,the multi-sensor target is matched by the exclusion method based on the characteristic distance and the Euclidean distance sorting algorithm;the matched target is fused by the Kalman weighted coefficient algorithm.In the multi-target tracking module,the time association method is used to complete the association of the target information before and after the time;an effective life cycle inspection framework is designed to complete the judgment of the target life state,including six life states: target formation,new target generation,target disappearance,target persistence,target tracking and clutter signal;the ID update strategy is used to complete the identification of the continuous target,tracking target and new target In the framework of effective life cycle,LSTM trajectory is introduced to solve the problem of trajectory prediction when the target is lost temporarily.The motion analysis algorithm includes target trajectory smoothing module,target position parameters and motion parameters calculation module.In the target trajectory smoothing module,the feature scaling of trajectory data is completed.By establishing the cubic trajectory fitting model and discussing the penalty term and loss function of the trajectory smoothing model,the L2 regularization and Huber error loss function are used to verify the trajectory smoothing effect of the model in the simulation scene.In the calculation module of target position parameters and motion parameters in the lane,the calculation of target position parameters is completed by establishing the calculation method of target offset and heading angle in the lane where the target is located.The lateral velocity and longitudinal velocity of motion parameters are calculated by differential and exponential weighted average method of position parameters,and the model verification in the simulation scene is completed.