Research On Matching Based Platform Location And Target Tracking Method By Large Inclined Angle Imaging

S Platform positioning and target tracking are two important applications of scene matching technology in practice,which have great application value both in military and civil fields.In this thesis,the method of improving platform positioning and target tracking accuracy under large inclined angle imaging conditions is studied.Firstly,the similarity measure to adapt to geometric distortion and the method to estimate the confidence of matching result are studied to improve the matching performance.Then the methods are applied to platform positioning and target tracking under the conditions of large inclined angle imaging.The main contents are as follows:(1)In template matching,the traditional similarity measure can not be applied under the condition of image geometric distortion,and the similarity measure is proposed to keep invariant under the condition of similarity transformation.Using the bidirectional matching points in the image,the similarity transformation parameters between the images are voted,and the maximum number of votes is used as the similarity measure between the two images.Then,the voting lookup table and parallel acceleration strategy are used to improve the matching time performance.By means of interval voting,the similarity measure has certain adaptability to affine transformation and even perspective transformation.The experimental results show that the template matching can effectively reduce effection of geometric distortion and improve the template matching performance.(2)Aiming at the requirement of measuring the credibility of matching results in scene matching applications,a method of confidence estimation based on the combination of real-time graph and correlation surface features is proposed.Confidence with theoretical estimation method,the real-time image statistical features and correlation surface features constitute a feature vector,using random forest classification,decision tree,to support the correct ratio of matching results of confidence.The correlation surface multi peak coefficient is proposed to describe the ratio of the area in the correlation surface similar to the matching result,so as to represent the influence of the repeating pattern in the reference map.A large number of simulation images are used for matching and confidence estimation.The results show that using the matching confidence calculated by this method to classify the matching results correctly,the performance is obviously improved compared with the direct use correlation and the theoretical estimation.(3)Aiming at the requirement of high precision positioning in SAR / INS integrated navigation,a SAR platform localization method combining high precision scene matching and INS information is proposed.The SAR image is used to estimate the position of SAR platform at the imaging time,and the INS data is corrected by a sequence of the estimated position.The accuracy of the SAR platform is improved effectively.According to the error transmission relation,the influence of each error factor is analyzed,and the formula of platform positioning accuracy is deduced.A large number of simulation results show that this method is feasible and the error analysis is correct.The experimental results show that the proposed method can effectively reduce the drift error of inertial navigation system,and can realize high precision SAR platform location.(4)A target tracking method based on structure from motion is proposed to meet the requirement of high precision tracking for measurement of target in complex stereo scene.The relative motion trajectory of the camera is estimated by using the auxiliary feature points in the image,and the estimation of the target image position is obtained to improve the tracking efficiency.In particular,aiming at the long-distance imaging conditions,it is proposed to initialize the feature points using the weak perspective imaging model,to improve the tracking stability with the relative attitude angle is used as input,and to decrese the drift error of the image prediction position the target space position update strategy with the auxiliary feature points is proposed.Simulation results show that the proposed method can effectively improve the target image position prediction accuracy,and ensure the target tracking performance.The tracking results of real sequence images show that this method can realize target tracking under the condition of flying around,and keep good prediction performance when the target is out of view or occlusion,then quickly recapture the target.In the following work,further research is needed to develop method and technology to improve the real-time performance of platform positioning and target tracking.