Study On Multi-Source Method For Moving Target Detection In Complex Low Altitude Environment

Detecting techniques of moving objects based on multi-source cooperation plays a significant role in security monitoring,emergency searching and rescuing,avoiding risk in lower airspace,and so on.As for security monitoring,with the increasingly widespread application of unmanned aerial vehicle(UAV),the corresponding supervision has not yet kept pace with the development of the vehicles.UAV is a typical moving object with features of low radar cross section(RCS),small size,and slow velocity.Thus,it is not reliable to realize the dependable detection and tracking of objects through obtaining data of UAV from multi-source sensors followed by data processing.For emergency searching and rescuing,it is also difficult to get the data containing weak moving objects and strong clutter background from the multi-source sensors anchored on the lower airspace moving platform,and then to separate the weak moving objects from the strong clutter background.Therefore,it is essential to research the detection of moving objects based on multi-source cooperation,whose technical breakthrough will make sense for improving the ability of security monitoring of low RCS,small size,and slow moving objects,and for enhancing the capability of all-time,all-weather searching and rescuing under complicated cartographic background.For monitoring UAV with low RCS,small size,and slow velocity,to address the limitations of traditional methods,including the high complexity of computation,inferior real-time effect,and the unreliable detection,this thesis employed the sparsity of objects and the low rank of background to separate the objects with high dimensional data set from the background via matrix decomposition and dimension reduction.Besides,we also utilized DECOLOR,RPCA,Godec,and MAMR algorithms to process and verify the measured infrared data flow and the optical data flow,respectively.The results suggest that DECOLOR method could relatively completely detect both large and small objects other while consuming a long time.Although expending the shortest time,the conventional RPCA algorithm could only detect bits of point set that the shape of moving object cannot be completely acquired.Furthermore,for large objects,the RPCA approach also generated cavity and leaved out profile.Godec algorithm is better than its RPCA counterpart in terms of detection effect,while the similar shortcoming still exists.MAMR method improved the missing profile and the cavity,whereas the obtaining the complete shape of small moving objects still remained untouchable.Aiming at the issues of VideoSAR containing low signal-to-noise ratio(SNR)caused by the large velocity of moving objects,and weakened ability of eliminating the strong clutter background resulted from the low accuracy of time-space matching between frames,this thesis carried out the research of shadow detection of moving objects.The difficulty of directly defocusing detection in the imaging of rapidly moving objects makes the shadow an important technique to detect object.There are problems of excess false alarm targets and difficulty in eliminating static shadow by directly applying the traditional stationary shadow detection model into the VideoSAR moving object detection.With the help of the time-space relevance between frames,we proposed a VideoSAR-based moving-object detection method established onto the time-space relevance.Firstly,we employed the basic static shadow detection module to process the single-frame image from VideoSAR,and got the preliminary shadow object,including both moving shadows namely objects and static shadows like trees and low-lying areas.To enhance the detection efficiency and lower the detection threshold,we collected shadowed sets with excess false alarm targets.Then,to reduce the false alarm,we adopted the time-space relevance method to implement the track-before-detect for shadowed sets obtained from the static model detection,namely utilizing clutter shadow ratio to estimate the motion parameter of objects.After that,the obtained motion parameter was used to conduct the Kalman filtering for the shadow,and thus screened out the shadow following the motion trail.Finally,the characteristic of the sliding window was used to determine the distance between centroids between the initial and the final frames among the group.The shadow is a static one if the distance between centroids is less than the threshold value;otherwise the shadow will be the shadow of moving object,achieving the elimination of static shadows from the moving object sets with large probability.The proposed arithmetic enhances the accuracy of VideoSAR moving object detection,and verifies the availability of this arithmetic through processing the measured VideoSAR data from the Sandia laboratory.