Research On Target Tracking Technology In Dynamic Uncertain Battlefield Environment

With the development of science and technology,the form of warfare in modern warfare has changed with more diverse types of weapons and equipment,more complex battlefield environments,and more massive information capacity.Among them,the human brain alone is responsible for the information in the battlefield.Processing has withdrawn from the stage of history.How to gain information advantage faster can take the initiative in the battlefield.With the rise of artificial intelligence and deep learning technology in recent years,computer vision represented by target tracking has been further developed,and target tracking technology can be used in the battlefield environment to real-time enemy targets Tracking,gaining the battlefield situation,so as to assist the commander in decision-making,and then win the battlefield initiative.Different from the ordinary daily environment,the battlefield environment is more complex and multi-end,including the target shape and movement state is uncertain,and the real-time performance is high;the target and the field of view have uncertain dynamic motion;the number of targets,the time of appearance and disappearance,and the location information are different.Confirmed;the target’s trajectory and identity are uncertain;the disturbing factors of the background environment in the field of view(such as explosions,smoke,etc.)are uncertain.In response to the above problems,this paper selects the most severely deformed human-shaped target in the battlefield environment,and conducts research on the single-target tracking task and the multi-target tracking task in the battlefield environment.The main research contents are as follows:1.Build a battlefield environment data setBased on the lack of data sets currently used in the battlefield environment for target tracking tests,this article builds a single target tracking data set based on the OTB-100 single target tracking data set and the MOT Challenge 16 multi-target tracking data set.Target and multi-target tracking battlefield environment data set to build a simulated battlefield environment and use it for testing.2.Technical research on single target tracking method in battlefield environmentA single important target that appears in the battlefield environment often needs to be tracked.However,because the tracking process is often affected by background interference,motion blur,occlusion,and deformation,an improved correlation filter tracking algorithm based on BACF is proposed.First,the extraction of color attribute features is increased to improve the resolution of the target;second,the model update strategy of multi-peak detection is used to suppress the interference of similar features;finally,the ability to extract target feature response maps of different scales is increased,and the scale is reduced.The interference of change was finally tested on the public data set and the battlefield data set.3.Research on multi-target tracking method technology in battlefield environmentWhen tracking multiple targets appearing in the battlefield environment,there are often the characteristics of dynamic background changes,high complexity,serious target deformation,and uncertain target appearance and disappearance.In order to ensure the accuracy of tracking and real-time requirements at the same time,select improvements The DEEP SORT algorithm first uses the improved YOLOv3 detector module based on ShuffleNetv2 to detect targets in the field of view to improve the real-time performance of the algorithm;on the basis of Kalman filtering,an improved fusion tracking tracker with MeanShift is proposed to improve the position of the target Then the color feature information is added on the basis of data association,and the data is correlated with the intersection ratio and area weighting to improve the accuracy of the association;finally,the test is carried out under the public data set and the battlefield data set.The experimental results on the public data set show that in the face of single target tracking in the battlefield environment,the accuracy of the algorithm in this paper is impro ved by 1.3%compared with the BACF algorithm,and the success rate is increased by 1.4%,which is greater than other classic related filtering algorithms..Tested on battlefield environment data,the tracking success rate and accuracy are both over 80%,and it has a good effect on the single target tracking problem in the complex battlefield environment;in the case of multi-target tracking in the battlefield environment,public data The algorithm in this paper is 65.2%and 82.9%in MOTA and MOTP indicators respectively.In the battlefield environment data set,the average accuracy of the algorithm on the data test is 63.9%,and the average accuracy is 83.1%.In the face of complex battlefield environments,With regard to target tracking,it has a good effect.