Maneuvering Target Tracking And Self-organizing Sensor Networks

Target tracking is one of the basic subsistence of the life. As a technological field studied for many years, it has been applied in a lot of fields such as traffic and military areas widely. The current researches of the target tracking are mainly concentrated on the maneuvering target tracking and multisensor data fusion.The main works of this thesis are following:I. Single Sensor Single Maneuvering Target Tracking1. Based on the constant velocity model, the principle of how to adjust the process noise covariance matrix Q is analyzed. According to the fact that the innovation has bias when the target is maneuvering, an adaptive filtering algorithm based on the innovation bias is given. It has the tracking performance of multi-model algorithm with the single model computation.2. Based on the turn rate model, a turn rate estimation method is designed. By analyzing the model set structure design of the Adaptive Interacting Multiple Models (AIMM), the design methods of the model set size, the distance between the models and the model transition probability are given. According to these methods, algorithms with better adaptive abilities than the conventional AIMM is designed.II. Multisensor Multitarget Tracking1. A new geometric multisensor data association method according to the centralizing level of sensors' line of sight is given. The algebraic description of the centralizing level is formed. The smallest sensor numbers guaranteeing the correct association in the multitarget and missing detection environment is analyzed. With the improved Genetic Algorithm, an all-neighbor estimation algorithm with and without the target state transition probability is designed.2. In order to limit the sensor numbers of the crossing location association method in high clutter or target density environment, a generalized likelihood ratio approach to eliminating ghosts is given. The approach is based on the fact that the sound intensity distribution of the same target is identical.3. Based on the Unattended Grounded Sensor networks, the sensor configuration for a static target and moving targets are discussed, which show that sensors distributed uniformly on a circle is suitable for sensor networks initialization. An adaptive Hopfield network with adjustable neuron thresholds is given, which can be used to form a target tracker self-organizingly. The approach can get tracking precision well enough with sensor numbers as small as possible in realtime.All of the algorithms above can be combined as a maneuvering target tracking system on the self-organizing sensor networks.