A Research On Tracking Algorithm Of Multi-sensor Data Fusion Based On Interactive Multiple Models (IMM)

Regarding the environmental complexity, randomness and diversity of the tracking objects, a study on dynamic object tracking is a topic not only full of challenges in both theoretic aspects and its applications, but also with great implications on further theoretic discussions and engineering applications. Due to mobility feature of the dynamic objects, the articles focused on the research and discussion for the tracking algorithm of multi-sensor data fusion based on interactive multiple models (IMM). Considering relatively inaccurate state estimation of dynamic objects obtained from the use of a single sensor, multi-sensor data fusion was a method used more frequently for a better state estimation of dynamic objects. Therefore, multi-sensor data fusion was the focus of this research.The first part of the article introduced the current global situation of research on the theory of dynamic object tracking. Some basic problems on dynamic object tracking were described first, followed by detailed elaboration on the algorithm of interactive multiple models. The article analyzed the basic theory of Fixed Structure Multiple Modes, and then pointed out its limitations. The research concentrated on the algorithm of interactive multiple models, which was considered as the highest cost effective hybrid estimation algorithm.In the next part, the research used probability hypothesis density (PHD) filter as the first-order statistical moment in the Bayesian spam filtering models. Using this method, the posterior probability density of random set of the dynamic objects state was unnecessarily calculated for every filtering moment, thus relieving the computational burden. At the same time, as the use of multiple models was appropriate to estimate the state of the time-varying system,'the algorithm of interactive multiple models (IMM) was hereby integrated with probability hypothesis density (PHD) filtering paradigm, becoming a news combined filtering algorithm, named IMM-PHD. IMM-PHD filtering covered the both advantages of tracking dynamic objects and data fusion from IMM and random set theory respectively, resulting in a better tracking effect. As sequential Monte Carlo sampling method could enable the operation of filter under a nonlinear/non-Gaussian condition, the filtering was done by sequential Monte Carlo sampling method. A comparison was then made between IMM-KF filter and IMM-PHD filter by using MATLAB simulation, the experimental outcome proofed that IMM-PHD filtering algorithm was more effective on tracking dynamic objects.In the final part of the article, IMM-PHD algorithm was implemented through radar/infrared sequential filtering methodology. Date observed by infrared was filtrated directly by IMM-PHD particle filter, then outcome yielded by infrared would set as the predicted value for the radar detector. The alternative function of radar and infrared ensured the visibility of the object state, thus the output of the filter would reveal the information on the object state. The feasibility and the effectiveness of the algorithm were then verified by MATLAB simulation.