| In midcourse defense, infrared seeker is employed for the precise homing guidance of interceptor, ensuring that the interceptor can kill the incoming target successfully. Multi-target tracking is an important part of information processing for the infrared-based interceptor homing guidance, and it is one of the key technologies to ensure the success of the interception. In distant imaging, incoming targets appear to be point targets in the image during most part of the time, so the tracking mainly depends on the dynamic information of targets. In the intercepting flight, interceptor involves drastic ego-motion, which is coupled to the infrared image and leads to the oscillation that target postions suffer. In practice, the inertial measurement equipment is usually installed on the interceptor following a strap-down manner, and it can measure the real-time attitude angle of imaging platform, which can be used to compensate the oscillation motion of target. This thesis studies on the multi-target tracking technology for homing guidance based on infrared strap-down platform, and focuses on the problems of maneuvering target motion, changing target number, the constraint of computational resource, infrared clutter disturbance and so on. The main work of this thesis is as follow:1) Proposes the motion and measurement model for midcourse target tracking that are based on the strap-down platform. In high-speed intercepting flight, the midcourse target appears to be maneuvering in the infrared image along with the target range changing. To accommodate the target maneuvering, this thesis proposes an adaptive "current" statistics(CS) model for the motion model of midcourse target. The proposed method uses online information to set the maneuvering frequency and acceleration variance of the CS model, which can improve the accuracy of the target acceleration estimation, and characterize the motion of midcourse target more effectively. Moreover, the measurement model turns out to be complex when introducing the inertial measurement to compensate the platform ego-motion. Then, basing on the analysis of imaging measurement error, inertial measurement error and decoupling error, this thesis proposes a simplified method, which ignores the inertial measurement error and decoupling error. The simulation results show that the proposed target model and measurement model can be used to track the midcourse target accurately.2) Fast multi-target tracking in the midcourse random scenario. Under the conditions of the dense targets and changing target number, the data association based on traditional tracking methods involve the deterioration problems on tracking accuracy and computational efficiency, which does not meet the requirements of homing guidance based on strap-down platform. In recent years, finite set statistics(FISST) has become the most important tracking theory. FISST outperforms traditional methods in terms of accuracy and efficiency, and shows a promising application in the rapid multi-target tracking for midcourse targets. This thesis builds the basic framework for target tracking based on the Gaussian mixture(GM) implement of linear complexity-cardinalized probability hypothesis density(LC-CPHD) filter, i.e. GM-LC-CPHD filter. The conventional birth intensity for the GM-LC-CPHD filter needs to search newborn target in the entire area, which leads computional inefficiency. This thesis proposes a novel method to adaptively build the birth intensity using the measurments. The proposed method uses the M/ N test approach to accumulate the information of multiple infrared image frames, and obtains the initial states of newborn target automatically, which can reduce the search area and improve the computational efficiency. Furthermore, the results of the conventional GM-LC-CPHD filter do not contain the target identity information, so the estimates can not be formulated as continuous tracks for each target, which can not meet the requirement for the object recognition module. This thesis introduces a labeling method for the track management of the GM-LC-CPHD filter, and proposes an improved approach for Gaussian elements pruning and mergence, which can output the tracks of each target more robustly and rapidly. Finally, the simulation results show that the proposed multi-target tracking method can achieve accurate and efficient multi-target tracking in the random midcourse scenario.3) Robust midcourse multi-target tracking under dense infrared clutter. Dense infrared clutter tends to cause error results in target tracking. This thesis proposes a method to use amplitude information(AI) to suppress clutter in GM-AI-LC-CPHD filtering for multi-target tracking. Firstly, the proposed method uses the Gaussian distribution to model the amplitude of midcourse infrared point target. And under the condtion of unknown Signal-to-Noise Ratio(SNR), this thesis proposes to use the latest L frames measurements to estimate the target amplitude and detection probability. Then this thesis derives the new recursions for the amplitude-aid GM-AI-LC-CPHD filter, and uses the amplitude-aid weights to distinguish targets from clutter in filtering. The simulation results show the proposed GM-AI-LC-CPHD filter can effectively reduce the probability of error tracking. |
