Research On Target Perception Technology Based On Unmanned Moving Platform

The unmanned moving platform is an integrated system which integrates many functions such as environmental sensing,dynamic decision and planning,behavior control and execution.Environmental perception is very important of the study of unmanned platform systems,and it is the premise and basis for decision-making and planning of the unmanned platform.Therefore,the detection system becomes an indispensable part of unmanned platform.In these detection systems,the infrared imaging detection system is loaded by most unmanned moving platforms,because it has the sensing of the surrounding environment,anti-jamming,all-day work and other advantages.So the target detection and tracking based on infrared image sequences has become one of the key tasks of unmanned platform.This paper focuses on the research of the infrared small target detection and infrared area target tracking based on the unmanned platform,designing and implementing an infrared target detection and tracking system based on unmanned platform.The main research contents and results as follows:(1)Aiming at the detect difficult problem of infrared small target detection in unmanned platform,a single frame infrared dim target detection method based on variational Bayesian Patch is proposed.First,the infrared image is decomposed by using the sliding window to obtain a number of Patches;Second,the k-Nearest-Neighbor(k NN)method will be using to divide all small areas into clusters of patches,and then using the de-mean method to remove the background components contained in each cluster;Finally,through the threshold segmentation method,so as to achieve the purpose of small targets detection.The simulation results show that the proposed method can effectively detect the small infrared target,and has a stable ability to suppress the background.(2)In order to solve the problem of the TLD which can’t be stable and continuous tracking target,this paper proposes an improved TLD tracking method based on the multi-scale feature of compressed domain.Firstly,the tracking module obtains the multi-scale feature of the image by using the rectangular window of different size,and uses the compression sensing method to reduce the dimension of the multi-scale feature,and then obtain the target position by the naive Bayesian classifier;Secondly,the detection module detects each frame of the image and determines whether the tracking module tracks the target;Finally,the learning module evaluates the errors of the detection module based on the result of the tracking module,and generates training samples based on the evaluation result to update the target model of the detection module,at the same time,update the classifier of the tracking module to avoid similar mistakes in the future.The experimental results show that this method can realize the continuous and stable tracking,and improve the running speed of the original TLD tracking method.(3)Based on the method of infrared small target detection and the tracking of surface targets,this paper designs and implements a set of target detecting software based on unmanned moving platform.When the target is far away from the platform,the performance of the target in the image is relatively weak,and the detection method of infrared small target is used to detect it.When the target has a certain shape,then use the surface tracking method to track the target.The results show that the system can well realize the function of small target detection and surface tracking.