Track Estimation And Target Detection Based On Multi-Sensor Information Fusion

For the past few years,electronic technology and interference technology are developing very fast,the electromagnetic environment in the air is no longer pure.The track information measured by a single sensor can not meet the needs of modern warfare.Multi-sensor track fusion has become a research hotspot in recent years under the traction of military requirements.Through the correlation and fusion of the local track information obtained by multiple sensors,multi-sensor track fusion can obtain more accurate track information,improve the detection performance,and enhance the robustness and reliability of the system.First of all,this paper studies and analyzes the target motion models commonly used in the field of track fusion,and analyzes the principle of common target tracking filtering algorithms.This paper introduces the necessary coordinate transformation and time synchronization methods before track fusion,expounds the concept and method of information fusion,and introduces the structure of information fusion from two aspects:the physical structure of information fusion system and the level of information fusion.Secondly,the technologies of track association and track fusion are analyzed and studied,and the weighted statistical distance method,the modified weighted statistical distance method and the track association algorithm based on genetic algorithm are theoretically analyzed,and on this basis,the genetic algorithm is improved.The effectiveness of the track association algorithm based on the improved genetic algorithm is verified by simulation experiments.Based on the theoretical analysis of three commonly used track fusion algorithms,a track fusion algorithm based on time convolution network is proposed.The simulation results show that the comprehensive position estimation accuracy of the track fusion algorithm based on time convolution network is 10.96% higher than that of the optimal distributed track fusion algorithm without feedback,and the comprehensive speed estimation accuracy is improved by6.32%.The effectiveness of the proposed algorithm is verified.Finally,the single source and multi-frame information fusion technology based on vision is studied.In order to solve the problems of fast moving speed of airborne platform,complex target background and low detection rate of small targets,an air target detection algorithm based on multi-frame feature aggregation is proposed based on YOLOv3 network and Flow Net optical flow estimation network.The algorithm aggregates the feature images extracted from different frames to the current frame through the optical flow field,so as to improve the feature quality of the current frame.It can effectively improve the target detection accuracy of the model on degraded frames(motion blur,target occlusion,defocus,etc.),and through multi-frame feature fusion,the stability of the algorithm for target position and category detection is effectively improved.The experimental results with the refueling cone sleeve as the detection target show that the average detection accuracy of the algorithm is 9.87% higher than that of the YOLOv3 algorithm,the recall rate is 8.06% higher than that of the YOLOv3 algorithm,and the stability of the detection result is 20.14% higher than that of the YOLOv3 algorithm,and the algorithm reduces the probability of false detection and missed detection.It meets the requirements of the target detection task on the airborne platform for the stability of detection accuracy.