| In the intelligent driving system,the vehicle should have the functions of environmental perception,intelligent decision-making,autonomous driving,and so on.It has very important application value in the fields of transportation,scientific research,military,agriculture and other fields in the future.However,the intelligent driving vehicle’s perception of the target is the basis for the realization of the system function,and it is very important to effectively locate and track the surrounding pedestrians,vehicles and other objects by the sensor,so as to reduce the traffic accident rate and improve the driving comfort effectively.However,the traditional method of target sensing mostly uses the fusion of infrared,vision and other sensor data,the process is complex,the intelligent vehicle moves at high speed and there are many uncertainties in the actual scene resulting in a decrease in the target sensing accuracy and neglecting the target multiple scattering point model.Therefore,this paper simplifies the fusion process from the perspective of probability based on the Bayesian framework,and proposes cooperative sensing method for intelligent vehicle driving,which effectively enhances the effect of target location and tracking.This is of great value and application prospect.Because the intelligent vehicles equipped with sensors of different types with different types of observations,the process of the fusion algorithm complexity by using the general data.Then we propose a dynamic non-parametric belief propagation(DNBP)algorithm based on Bayesian framework in this paper.In this method,states of the vehicles and target can be expressed as a probability density function(PDF)to simplify the data fusion process effectively,and the estimated PDFs of target location and tracks are also available in non-Gaussian and nonlinear case because of the characteristics of the NBP.Due to the presence of occlusion and the blind leads to the decrease of the positioning accuracy of the bicycle problem in the intelligent driving scene,the “host”vehicle sometimes can’t obtain the measurement to the target which may reduce the target estimation accuracy,or even makes it totally failed.Therefore on the basis of the previously proposed method,we propose a multi-vehicle cooperative position sensing method to provide the fusion center with other vehicles measurement,which adds extra prior knowledge to enhance target sensing accuracy.Furthermore,according to the uncertainty of different vehicles,a weighted vehicle cooperative scheduling strategy is proposed,and the perception accuracy in proposed method is significantly stronger than the equal gain of multi-vehicle collaborative sensing accuracy which further meet the requirements of intelligent driving precision.On the other hand,for the extended target multi-scatter model,the actual existence of the contour causes each scattering center to cause different observations and the assumption that the targets are usually considered as point targets is no longer valid.Then in this paper,we analyzes the Bayesian estimation approach to extended target tracking based on random matrices,which can estimate the kinematic state and the contour characteristics of the extended target simultaneously which can be applied to vehicle tracking scene effectively.This paper will be carried out for the above research,and verify the feasibility of the conclusions obtained through the experimental simulation,in order to provide theoretical support for the effective application of the algorithm is conducive to follow-up to further improve the accuracy of point-target sensing by introducing the current hot technology,as well as the actual case maneuver expansion goals and other aspects of in-depth study. |
