Font Size: a A A

Contour And Edge-based Visual Tracking Of Non-cooperative Space Targets

Posted on:2014-01-18Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y LiFull Text:PDF
GTID:1108330479979646Subject:Aeronautical and Astronautical Science and Technology
Abstract/Summary:PDF Full Text Request
Accurate measurement of the relative position and attitude, or pose in general, between spacecrafts is the precondition for realizing space missions such as space rendezvous and docking, space attack-defense and spacecraft on-orbit servicing. Optical imaging sensor based vision measuring technology utilizes optical imaging sensors installed on the chaser spacecraft to image the target spacecraft, then extract its image features and calculate to conduct pose measurement. Compared with other measuring technologies, optical imaging sensor based vision measuring technology is more intuitive and autonomous, and can achieve high level of precision, it is the main relative pose measuring means in final approaching phase for the space operations mentioned above.Target spacecrafts can be classified into cooperative targets and non-cooperative targets depending on whether cooperative markers are installed or not. Vision measurement for cooperative space targets has been studied in depth and applied successively. Relatively speaking, it is full of challenges to conduct vision measurement for non-cooperative space targets. It has more universal significance since most of the space targets are non-cooperative, and it is gradually becoming a new research hotspot.Full advantage has to be taken of the inherent features of non-cooperative space targets for there are no cooperative markers. Contour and edge are two important kinds of image feature, contour is the closed boundary which surrounds target region in image, and edge is the image region where gray scale, colour or texture acute is discontinuous or suffers acute changes. As for long distance object, its imaging size is relatively small, detailed features are difficult to be extracted, but its contour is still available, so long distance pose estimation can be done based on contour. As for short distance object, its imaging size is relatively big, compared with other features, edge extraction is more fast and more robust to illumination variation, it possesses a big advantage to realize pose estimation through edge features.Visual tracking methods for non-cooperative space targets based on contour and edge are studied in depth in this dissertation, including two major parts: 1. Contour extraction and tracking methods, which are 2D tracking essentially, for long distance targets; 2. Pose estimation and tracking methods, which are 3D tracking essentially, for short distance targets. The result of contour extraction and tracking can be input to the existing contour based pose estimation methods for long distance pose estimation, besides, it can also provide initial values for accurate short distance pose estimation and tracking. In general, the innovativeness of this dissertation is embodied in following points.In the aspect of target contour extraction and tracking.1. Lines Grouping and Saliency Analysing based Perceptual Organizational Contour Extraction method, LSPC, is proposed.Conventional contour extraction methods cannot eliminate the cluttered image background when it is complex, while perceptual organization based contours grouping can handle this problem effectively, for it does not need a priori knowledge, and it can organize only according to orderliness among image features. Firstly, LSPC method transforms the image segments, connections, and connecting costs into vertices, arcs, and weights of arcs in a weighted directed graph. The weights of arcs are quantized according to proximity, continuity and similarity of perceptual organization Gestalt Law. Then the problem of contour extraction is transformed into the problem of finding the shortest path in a weighted directed graph. Finally, the most significant contour is found through effectiveness validation and saliency analysis. Significant contours of target can be rapidly extracted using this method.2. Motion Probability based Geometric Active Contours method, MGAC, is proposed.A contour extraction method combining motion segmentation and GAC is proposed. From the point of view of motion segmentation, MGAC first compute KLT feature optical flow and conduct clustering and motion model estimation simultaneously, then computes the probability of each pixel belonging to the target, and extracts contours using geometrical active contour. Contours of mobile target under complex background can be steadily extracted using this method.3. Histogram based Geometric Active Contours method, HGAC, is proposed.After the contour of target is extracted, HGAC first calculates the color histogram of the interior zone of the extracted contour, and take it as a target template, then evolve the contour according to principle “Maximizes the similarity level between the interior zone of contours and the target template, and Minimizes the similarity level between the exterior zone of contours and the target template”, and finally realize stable tracking of target contours.The results of contour extraction can be used not only for pose estimation but also for feature extraction, morphological description and identification, etc.In the aspect of model based pose estimation and tracking.1. Normal Distance Iterative Reweighted Least Squares, ND-IRLS, and Distance Image Iterative Least Squares, DI-ILS, pose estimation methods are proposed.ND-IRLS can cope with the problem that ND-ILS is susceptible to interference of noise and background by weighting the samples. DI-ILS has better realtime performance by simplifying measuring procedure. These two methods are both computationally efficient, robust to certain extent of initial guess error, and possess high accuracy.2. Least Squares embedded Particle Filter, LSPF, pose estimation method is proposed.Particle filtering is a commonly used Bayesian filtering technique, LSPF introduces the Least Squares Optimization method into PF frame by importance sampling technique, and combines the advantages of high-precision, high level of realtime performance of Least Squares method and the robustness of PF together. This method only requires a small amount of particles to perform stable and accurate object tracking.Besides, specific working flow design of edge model based pose estimation and tracking methods is studied in detail in this dissertation, researches are conducted in several links, such as model preparation, orientation change prediction and multiple cameras measurement, etc.This dissertation is a good attempt to visual tracking for non-cooperative space objects, and also a good supplement to image interpretation of shooting range. And the achievements has been applied successfully to 3D pose estimation and tracking experiment of space non-coorperative target and 2D pose interpretation of missile target lauching experiments. Besides, the studies of this dissertation are expected to be applied further in other vision-based measurement task, such as vision-assisted landing of aircraft and scene matching, etc.
Keywords/Search Tags:Non-Cooperative Targets, Visual Measurement, Contour Extraction and Tracking, Edge Registration, Model-based Pose Estimation and Tracking
PDF Full Text Request
Related items