Jinchun Wang

Attitude determination using a single-star sensor and a star-density table

An alternative approach to attitude determination using a star sensor is presented. Whereas conventional star trackers require star vector observations through an identification of star constellations and a tracking of the identified stars, the proposed method takes multiple vector observations of virtual lines of sight instead of stars. A virtual line of sight is the pointing direction of a small portion of a star sensors field of view and the vector to this line of sight is measured by searching celestial positions having the same theoretical star densities with the measurements, defined as the number of detected stars in the field of view. The suggested approach is based on the fact that the distribution of stars in the sky is not homogeneous. A stepwise search method to determine the pointing directions of multiple lines of sight that give the vector observation sequences is proposed, and a simple least-squares solution is applied for the attitude determination using these vector sequences. The proposed method allows not only a stabilized spacecraft but also a rotational one to take measurements; in the latter case the vector observations of stars are usually difficult owing to the sensors motion.

GIS-assisted image registration for an onboard IRST of a land vehicle

The target detection can be carried out with a statistical matched filter. The construction of the matched filter needs the information on the background clutter statistics as well as on the shape of the target. For the computational simplicity, a filter bank consisted with pre-designed matched filters can be used for adaptive filtering. The classification of background clutter must be preceded to compose a filter bank and need pre-collection of samples of background clutter. In land-based IRST, there are too many different types of background clutter to hold a filter bank tuned to them. To overcome this difficulty, we propose a new classification method which use GIS (Geographic Information Science) -assisted background registration. We discern different clutter regions in the initial image frame using a feature vector composed of the vertical and the horizontal autocorrelation and build filters tuned to each class. In the successive frames, we classify each region of different clutter from contour image obtained by projecting the GIS data and by registering to the previous image. Each classified region of image is then filtered using a pre-designed matched filter in the previous image frame. We only have to construct a filter for newly appeared region. The proposed algorithm has been tested with synthetic image frames, and we observe that our method has advantages of reducing computational load and false detection at edges.

Image registration for an imaging system on-board fast moving military vehicle

We present an image registration algorithm based on the area-correlation method. Image registration is indispensable for moving target detection and tracking with the infra-red search and tracker (IRST). In our approach to image registration, displacements between frames are obtained using the area correlation, and then, the search image is rectified to the reference image by compensating displacements. Detection of a moving target is achieved by the frame difference method. The proposed algorithm has been tested with real image data, and we observe that the proposed algorithm is able to compensate the displacement of two images in that we get only a few detection points when we subtract two registered images.

Extended target detection and tracking using the simultaneous attitude estimation

We propose a new extended-target detection algorithm which can adapt to the time-varying target shapes. We estimate the attitude of the target using an extended Kalman filter with a sequence of image frames. The estimated target attitude is then used to predict the projected shape of the target image. Using the predicted target shape, we can construct a better-tuned matched filter for the detection of the target in the next image frame. The proposed algorithm has been tested with synthetic IR image frames, and we observe that the false alarm rate has been reduced by the order of magnitude in comparison with the simple matched filtering method with the Gaussian-shaped target assumption.

Attitude determination of a spinning object using dual imaging sensors and a star catalog

A bootstrap filter algorithm, which uses a sequence of the number of stars to estimate the attitude of a spinning object, is presented. This choice of measurement makes the algorithm practicable to apply to the spinning object without any initial acquisition lock. The statistical model of measurement is derived and incorporated to make a star-density map. The conic fitting method is used to obtain measurements in the star sensor images. The simulation result is presented which demonstrates the ability of attitude estimation and the fast convergence property of the proposed algorithm.

Angular velocity estimation of fast spinning object using an imaging sensor

We propose a new spinning object angular velocity estimation method using an imaging sensor. Most of the previous algorithms require a knowledge of the attitude or of the body vector measurements. For a fast spinning object, however, the body vector measurement is a difficult task due to flows of the object image. To overcome this difficulty, we search for the angular velocity which minimizes the cost function defined by the Euclidean distance differences of the observed star traces and the derivative star traces of an estimated angular velocity. With this formulation, angular velocity determination becomes a kind of minimization problem in 3D of the angular velocity vector. To reduce the search dimensions and use search techniques in 1D, we randomly sample points on the star traces and these random points assign a constraint to the possible angular velocity vector according to the kinematics equation. A numerical simulation for fast spinning cases is carried out to establish the results.

Registration of background images using correlation for a land-based IRST system

Image registration is useful for the moving target detection and tracking in the infra-red search and trackers. We present an image registration method based on the maximum likelihood principle. With our approach to the image registration, displacements between frames are calculated using the correlation, and then, the new image is rectified to the reference image by compensating the displacement through two steps of coarse and fine rectifications.

Multiple IR target tracking in clutter environment using the Viterbi algorithm

We address the formation of a Viterbi algorithm for target tracking after detection. A target tracking after detection process can be made by a Kalman filter. The Kalman filter, however, may give some false tracks which are induced from false alarms. In this paper, we introduce an alternative approach to the target tracking based on the Viterbi algorithm. The state of the Viterbi algorithm includes the position and velocity of the target, and the measurement vector is the detected target position(in 2D). Because a target cannot maneuver abruptly due to its dynamical limitation, the velocity vector cannot be changed suddenly in direction as well as in magnitude. From this fact, we can define the state transition probability as a function of changes in angle and speed between the present state and the previous state. The proposed algorithm has been tested, and we observe that it tracks multiple targets accurately while the Kalman filter generates more tracks following clutter points. In addition, we have observed that a dynamic programming based approach fails to track the target.