Target tracking and navigation equipment such as the gps contain a Kalman filter which either acts to smooth data when the unit is unaided, or as an estimating filter when inertial data are accepted. The need to integrate multiple sensors results in a hybrid system with extremely large computational requirements for real-time applications. Often the hybrid system takes the form of a "cascaded" filter to ease the computational burden. Sensor data integration is often difficult due-to correlating the time of the measurements and the different measurement rates of the sensors. In this proposal, parallel Kalman filter architectures are optimized for hybrid systems consisting of multiple sensors to achieve improved performance and stability. The computational advantage of parallel processing minimizes measurement time sensitivity and data transfer over the bus in the systolic architectures proposed. The partitioned square root and upper diagonal kalman filter algorithms proposed are capable of multi-rate filtering via a unique decoupling of the predictor and corrector equations in the filter while maintaining optimal estimation. The paralleled processing techniques proposed can be extended to the more general case of distributed filtering for real-time navigation, target tracking and scene analysis.
