Research On Active Visual Scheduling Method Based On Belief Rule Base

With the increasingly complex road conditions and the increasing number of motor vehicles,traffic video surveillance systems are of great significance to traffic safety.Active vision is an important part of traffic monitoring.The active vision system is a method to control the camera's movement and adjust the camera parameters to achieve the best shooting effect for the target.The active vision dispatch method is to ensure the monitoring system.An important method for normal and efficient operation.Due to the particularity of the organizational structure and working environment of the monitoring system,its scheduling method is quite different from the conventional scheduling methods in the past.In the active visual surveillance system,the individual computing power,aperture size,storage capacity,and physical space of the cameras are different,resulting in large differences in active vision scheduling methods for different shapes and different application scenarios.Surveillance cameras are apt to be limited by the shooting range and the quality of their own shooting,which causes problems such as lack of key information and repeated shooting.Based on the above factors,when researching the scheduling method of active vision in the monitoring system,we must comprehensively consider the impact of different factors on the camera in order to comprehensively and objectively realize the research on the scheduling method of active vision.This thesis has done the following work for the scheduling problem:Aiming at the intelligent modeling of traffic scenes,by analyzing the participating elements of the traffic scenes,the main traffic participating elements obtained are abstracted to construct a structured traffic scene model.Taking the traffic scene around the island as an example,this thesis proposes a method for modeling traffic scenes using parameter constraints.A static scene is established using the area constraint method.The combination of circular curves,transition curves,and elliptic curves is used to simulate the vehicle in various situations.The trajectory model of driving on the road establishes the connection between the static scene and the dynamic traffic trajectory.Aiming at the dynamic relationship between the vehicle target's driving trajectory and multiple cameras,the shooting data feature simulation method was used to obtain the shooting angle and shooting distance characteristics of the camera cluster for the vehicle target.The experimental data proved that the simulation data was obtained and combined with the actual traffic scene The distribution and trend of the data are analyzed.This method has high application value in road traffic simulation data acquisition and method simulation direction.Aiming at the scheduling problem in the active visual surveillance system,in order to organize each camera in the camera cluster into a unified comparison framework,this article introduces Belief Rule Base(BRB)into the field of traffic monitoring and control for the first time.In order to obtain a uniform shooting quality level,a series of confidence rules reflecting the shooting quality status of the camera were constructed,and a confidence rule table was established by combining the experimental data distribution with expert knowledge.As one of the most advanced technologies in the field of modeling complex systems,BRB can effectively use qualitative expert knowledge and quantitative data sets,and can regularly describe knowledge with fuzzy uncertainty and probability uncertainty.The Evidential Reasoning(ER)method is used in the inference process.This method can turn the BRB inference process into a participatory and visible process,and the results are interpretable and traceable accordingly.In order to solve the problem of insufficient model accuracy caused by the defects of expert knowledge and the parameter optimization of BRB model,this thesis uses a constraint covariance matrix adaptation evolution strategy algorithm based on projected operation(P-CMA-ES).Solutions that do not satisfy the constraints are mapped back to the constraint space through projection operations.The experiment proves that the model has a good effect on the camera shooting quality evaluation,and solves the problems of low accuracy caused by the shortcomings of expert knowledge,insufficient training in the case of small samples,and the degree of fitting compared to other traditional types of methods.There has been a substantial improvement.By taking the camera with the highest shooting quality as the target-camera corresponding to the current target to complete the camera scheduling work,the application research of BRB in active vision scheduling is realized.Finally,an active visual scheduling software was made based on the above method.