Research On Infrared Multi-target Tracking Method Based On Feature Extraction And Data Association

Infrared multi-target tracking technology is widely used in both military and civilian fields.Due to its imaging and scene characteristics,infrared detection images often have problems such as low image resolution and serious background clutter,which makes infrared multi-target tracking a huge challenge.In view of the low tracking accuracy and poor stability of the existing infrared multi-target tracking technology,the accuracy and robustness of the multi-target tracking algorithm are improved by constructing a target multi-feature extraction method;by improving the data association method,the multi-target is further improved While tracking accuracy,the time complexity of the algorithm is reduced,so that the tracking algorithm has real-time requirements.The main work of this paper is as follows:(1)Aiming at the problems of high similarity of different infrared targets in the air,many background noise points and similar noise points with the target,the target tracking is difficult.An air infrared multi-target tracking algorithm based on multi-feature fusion and hierarchical data association is proposed.First,extract the target's apparent feature,motion feature,and scale feature based on the infrared target characteristics;secondly,design a preliminary association method based on the multiple feature information of the target to eliminate the interference of the misdetected target;then,in the second level of correlation,Targets are classified according to the size of the scale,and different feature combinations are used for different types of targets;finally,the Hungarian algorithm is used to allocate the detection results.Experiments show that this algorithm can effectively track infrared multi-targets under the background of the starry sky.(2)Aiming at the pedestrian target tracking challenges caused by the high similarity of different pedestrian targets in the infrared pedestrian scene and the easy occlusion of the target,an infrared pedestrian multi-target tracking algorithm based on multi-directional neighborhood and trajectory confidence is proposed.First,calculate the degree of overlap between targets according to the intersection ratio and initialize the trajectory information;secondly,construct the multi-directional neighborhood appearance and motion characteristics according to the pedestrian target detection results to reduce the impact of occlusion on target matching;then,calculate the data association and update the trajectory status according to the data association result;finally,based on the historical information of the trajectory and the current state information to construct the trajectory confidence and judge the reliability of the trajectory.Experimental results show that this method can not only better solve the problem of pedestrian occlusion,but also improve the accuracy of multi-target tracking and improve the speed of the algorithm.(3)Design and implement infrared multi-target tracking system.The system is developed based on Qt and includes three functional modules: infrared target detection,multi-target tracking,and multi-band image fusion.Among them,the target detection module contains a variety of detection algorithms,which can cope with the difficulties in target detection caused by different complex scenes;the multi-target tracking module includes two parts of trajectory initialization and trajectory estimation,which can effectively track multiple targets;in the multiband image fusion module Different fusion algorithms can meet the fusion needs of multiple detection scenarios.