Research And Implementation Of Robust Tracking Of Video Targets In Ground Unmanned Combat Systems

The unmanned ground combat system is a kind of highly intelligent weapon platform with functions of detection,search,rescue,transportation and fire strike.It is an advanced system which detects target in complex scenes and achieves robust tracking by using video information,and has important research significance and engineering value.Based on the problems of detecting and tracking of moving target in the unmanned ground combat system,the main research contents are as follows:Firstly,an object detection algorithm based on optimized ORB features and background-based compensation is proposed.The feature points are extracted by calculating the Hessian matrix.And the feature vectors are calculated by the BRIEF descriptor.Then,the detection of moving target is realized by combining background-based compensation and the frame differentiation method.The experimental results show that the algorithm can detect moving object accurately in dynamic background.Secondly,aiming at the problems in the environment without severe occlusion,an adaptive complementary learners with diversified color attributes for object tracking algorithm is proposed.The CN features is added on the basis of HOG.And the adaptive fusion and update of model is constructed by the quality of response map.The experimental results show that the algorithm can achieve object tracking accurately.Thirdly,aiming at the problems in the environment with severe occlusion,an object re-detection mechanism is added on the basis of the proposed tracking algorithm.The feature library which is used to evaluate the completeness of target is constructed by SURF operator.The target can be re-detected according to the constructed SVM classifier when it is lost.The experimental results show that the algorithm can deal with the situation of lossing in the environment with severe occlusion.And the long-term robust tracking of the target is achieved.In addition,the algorithm is transplanted to JN-IGH57 x embedded platform based on Linux system.And the experiment is completed in the actual scenes.The experimental results show that the proposed method can detect moving targets in complex scenes and achieve high robust tracking.