Research On Multi - Sensor System Registration Algorithm

Registration of multi-sensor systems is the main step and preprocess for sensor networks and integrated information fusion systems. In the military field,along with the proposal and development of the battle theory of incorporation of army, navy, air and outer space, and the usage of various type advanced sensors, information fusion of multi-sensor system became more and more important; in the civilian field, along with the development of the world economics, the requirement of multi-sensor system became more and more in the fields of traffic and transportation, environment surveillance and industrial manufacture; along with the development of computer tech, the real-time filter fusion could be applied widely.The dissertation focuses on registration algorithm study for multi-sensor systems, including the delayed measurement processing registration algorithm, the algorithm for heterogeneous information, the gridlock algorithm for measurements in clutter, and robust estimation for little scale uncertainty bias. The main contributions of this dissertation are summarized as follows:1. The registration algorithm for delayed data of asynchronous sensors is proposed. For the delayed measurements of high data rate sensor, lagged time threshold filtering method, and detailed steps of computation method, is proposed firstly. Combined with IMM (Interactive Multiple Model) and MMSE (Minimum Mean Square Error), IMM-MMSE negative-time update algorithm for delayed data is given. On the registration problem of the delayed, with lagged time threshold filtering method and IMM-MMSE negative-time update algorithm, IMM based multi-lag out-of-sequence-measurement temporal alignment technique, relative computing steps, flow chart and numeric simulation are presented. This algorithm is on-line and real-time, which sufficiently utilizes the delayed measurements, excludes too old data, improves the precision and computation efficiency. It solves the contradiction of the just-drop method that could lead to lose efficient information and letdown precision, and the easy-process method that could lead to depress computing efficiency and more requirements of computation sources.2. Registration algorithm for clutter measurements based on particle filter is presented. Traditional registration algorithms have hypothesis that data sequences are normal measurement of true target, but sequences in clutter environment including false detection from true targets or fake targets, missing detection of true targets, so the traditional could not be applicable. For measurement in clutter, the problem is described with FISST(Finite Random Set Statistics)firstly, basic Bayesian filter and PHD filter (Probability Hypothesis Density) are clarified, correspond evaluation methods are introduced,the clutter data’s registration algorithm’s implementation method-approximate particle filter is proposed. Numerical instance accordingly is carried out to verify its feasibility, which can deal with data gridlock in clutter rightly.3. Geometric transformation based registration algorithm in multi-sensor systems is proposed. The short traces formed by targets’measurement of multiple sensors are transformed from measurement reference system to screen coordinate firstly. On the condition of single target, short traces are processed by geometric transformation alignment, relative bias in orthogonal coordinate achieved, and measurements compensated with Cartesian bias, filtering and fusion executed lastly. For multiple targets the association problem must be considered. The short traces position grads association mean is presented, track pair attained after association, the relative bias estimation of each track pair calculated, weighted bias of each sensor figured out, bias compensation carried out, and etc. The flow chart of algorithm and its Monte Carlo numerical simulation are presented. This algorithm is a kind of off-line method, which is briefness, efficiency, high precision, suitable in low-rate sensors and weak-maneuver targets only.4. FISST based dissimilar information of multi-source registration algorithm of multi-sensor systems is proposed. For surveillance and track of special targets with super high-speed and super strong-maneuverability, it is hard to efficiently detect and track only by traditional sensors such as radar, intelligence information and expertise base etc should be needed, but there is no practicable mean for management of multi-source heterogeneous information’registration. The special targets’characteristics are introduced and analyzed firstly, FISST theory be clarified and imported, some kinds of heterogeneous information be expressed with FISST,focused targets’matching probability as registration index of heterogeneous information be proposed, and results be computed with Kalman evidence filter. Numerical simulation validated the feasibility of method proposed.