| The detection and tracking of non-cooperative targets(meaning that the target does not provide its own location information to the system)is a key foundation for digital,intelligent automated systems in many fields,including security and digital cities.Many monitoring methods can be used for non-cooperative target detection and tracking methods,of which radar is cost-effective due to its advantages of all-weather,mediumand long-range detection;cameras are the most widely used due to their conformity to human eye vision,rich texture information and low cost.The information from radar and camera video each has its own characteristics,so fusing radar and vision-based target recognition tracking can greatly improve the detection recognition tracking capability of the system.Heterogeneous sensor fusion suffers from different sampling times,different expressions and different focuses,and how to effectively fuse heterogeneous sensor data needs to be determined based on the actual situation.In this thesis,we design and implement a heterogeneous multi-source sensor target track fusion method with radar point target and image block target track fusion as the topic.The main work of this thesis is as follows:1.for the heterogeneous multi-source sensing to be processed in the radar point target and camera block target,camera block target and camera block target fusion needs,proposed a unified point target,block target body intersection algorithm.The algorithm constructs different area hulls according to sensor types,determines whether different sensors observe the same target by means of camera projection models and convex polyhedral collision detection,calculates the area hull intersection region on a correlation basis,and uses the 3D representation points of the intersection region as the fusion result.The performance of the algorithm is validated using a publicly available multi-camera multi-target dataset.The traditional radar IMM＿UKF target fusion scheme is used for radar point targets and radar point targets in heterogeneous multi-source sensing.2.A heterogeneous multi-source track fusion method containing block targets is designed based on the above algorithm.Two simulation scenarios containing multiple targets and heterogeneous multi-source sensors are built using Unreal Engine4,and the performance of the heterogeneous multi-source fusion algorithm is experimentally verified based on a distributed system at the cloud edge.In distributed system simulation experiments,the fusion of radar point target and camera block target trajectories resulted in an average accuracy improvement of 15.17% and a maximum improvement of 44.58%compared to the radar point target trajectory. |
