Research On Sensor Deployment Design And Tracking Technology For Roadside Visual Perception

In recent years,roadside perception systems have been extensively studied due to their wide sensing perspective and stability of sensing results.However,there are still many problems in roadside perception that need to be solved urgently.In terms of sensor model construction and analysis,there is a lack of modeling analysis of roadside perception sensors;in terms of sensor deployment configuration,there is still a lack of a complete set of optimization methods for roadside sensor deployment; in terms of environment perception and cognitive algorithms,for Research on multi-sensor tracking algorithms is still insufficient.Therefore,this paper focuses on the roadside perception system,and conduct research on the roadside perception system configuration design and multi-target tracking technology.First,for the special passive sensor of the camera,according to the camera pinhole model,the ground projection geometric model of the roadside camera is constructed to determine the camera coverage judgment conditions and perception range.For the object detection task,the relationship between the camera detection probability and the depth is analyzed,the camera perception model is constructed,and the actual maximum detection range of the camera is determined.Secondly,a general sensor deployment optimization method is proposed.Based on the camera model proposed in this paper,by discretizing the actual road and the pose of the sensor to be deployed,a multi-objective optimization problem with the lowest cost and maximizing the average detection probability is constructed to obtain the Pareto optimal solution for the final sensor deployment.Finally,based on the concept of random finite sets and Poisson Multi-Bernoulli mixture filtering theory,a dual-sensor Poisson Multi-Bernoulli mixture filtering fusion tracking method is proposed for the special form of dual-sensor fusion perception in roadside perception.we construct a new dual-sensor Poisson Multi-Bernoulli observation model and proved that under the action of this model,the Poisson Multi-Bernoulli mixture posterior probability density is still in the conjugate form In addition,through mathematical induction,it is proved that the dual-sensor observation model proposed in this paper can be extended from an independent single-sensor model.