Research On Dissimilar Sensor Data Fusion Methods

The technology of dissimilar sensors data fusion utilizes heterogeneous sensors to detect object and obtain the information from many respects, and obtain better state estimate than single sensor. Due to the tracking accuracy of sensors is reduced in dissimilar sensors tracking system. This thesis provides some methods. Main research work of this thesis as following:1. This thesis expatiates on the concept of data fusion, basic principle and development of data fusion. Besides, it introduces some basic methods of state estimation under different rules.2. In practical sensors tracking system, duing to the bad influences such as background environment and fault sensors, it leads to real time estimates measurement noise covariance in accord with the property of innovation, and makes modifying measurement covariance keep orthogonal properties of innovation, then influences the Kalman optimal gain to modify state prediction values to make the covariance of prediction error smallest. The simulation results show that the modified algorithm can restrain the bad influences of outliers and improve the tracking accuracy.3. Since the sensor missing detection of target tracking system of the dissimilar sensors data fusion, this thesis presents an algorithm of data fusion. The algorithm calculates the similarity of measurement value of arbitrary two sensors based on the measurement predictor of multi-sensor fusion to the confidence of each sensor measurement value, the integrated similarity of each sensor measurement value and other sensors measurement values based on the theories of probability source combination and characteristic vector of non-negative matrix is calculated to determine the weight of each sensor. This algorithm can adjust the real time fused weights of sensors, and efficiently eliminate the bad influences of sensor measurement values of missing detection on measurement fusion. Simulation results show that the algorithms can effectively solve the missing detection of the sensors, and increase the tracking accuracy.4. Aiming at the problem of setting in advance for limit of acceleration in "Current" statistical model, an improved maneuvering target tracking algorithm based on displacement prediction covariance is presented. It utilizes extended Kalman filtering to carry out dissimilar sensor fusion. This algorithm avoids the adverse influence of setting up the limit of acceleration in advance to state estimation, and the simulation result shows the accuracy of the maneuvering target tracking is obviously enhanced by this algorithm.5. The influence of variance of the presupposed mobile dissimilar sensors measured error in actual environment is analyzed, a new algorithm is presented, the algorithm estimates stationary radar and mobile IRI measurement noise covariance based on actual innovation covariance in real time. It utilizes distance, azimuth and pitching angle estimates values and their covariance to asynchronous fusion, and uses extended Kalman Filtering to obtain target state estimated value. The simulation result shows that algorithm can obviously improve the accuracy of the maneuvering target tracking.