Research On Accurate Target Tracking Method Based On Deep Fusion Of Heterogeneous Data

Target monitoring is an important part of urban smart security,and its perception accuracy also determines the in-depth application of smart security.The innovative introduction of millimeter-wave radar provides a new solution for the task of accurate target perception.However,due to the limitations of production costs and physical bottlenecks,millimeterwave radars have equipment performance shortcomings.In order to make up for this defect,the mainstream solution mainly adopts the multi-sensor fusion strategy to improve the perception accuracy.However,traditional multi-sensor fusion methods only simply combine the fused data,and do not adequately address the tracking accuracy in depth and lateral ranges.Therefore,this paper focuses on the problem of insufficient heterogeneous data fusion around the indoor moving target perception scene,and proposes a deep fusion filtering algorithm(fusion-EKF)based on millimeter-wave radar and camera,which realizes the deep fusion and crossover of heterogeneous data.It is verified that the effect of precise target positioning is achieved.On top of this,in order to speed up the processing speed of the perception system,a radar data preprocessing model is designed to improve the quality of the perception data,so as to meet the real-time requirements of the perception system.Aiming at the perception accuracy problems of poor angular resolution and low cross resolution caused by inherent defects of millimeter-wave radar hardware,a deep fusion algorithm based on heterogeneous data is proposed.The advantage of the camera equipment is used to correct the millimeter-wave radar data to improve the resolution and target recognition ability of the millimeter-wave radar.Completed deep fusion and cross-validation of heterogeneous data.Experiments show that the proposed fusion algorithm can achieve a distance accuracy of 0.29 m and an angle accuracy of 0.013 rad.Compared with the traditional fusion algorithm,the target perception accuracy is improved by 15%.According to the real-time requirements of the target perception system,the smooth drawing of the target trajectory is realized.In this paper,combined with practical application scenarios,a radar data preprocessing model is proposed,which defines the invalid data in the original radar detection data and completes the filtering,which reduces the occupation of system computing resources by invalid signals and improves the real-time performance of the system.Experiments show that the proposed preprocessing model can reduce the amount of radar data by 30% without affecting the target detection accuracy.Based on the researched algorithms and schemes,a building visualization system is designed and realized by relying on the micro-Io T sharing platform,and the construction of the proposed fusion model is completed by accessing the sensing equipment and calibrating it in space and time.The system,real-time performance and accuracy of target location tracking are verified in performance tests of multiple indoor personnel detection scenarios.The proposed algorithm has been applied to building security,smart sentinel and other systems and has achieved certain commercial value.