Research On A Posterior Laplace Approximation Direct Tracking Algorithm Based On The Improved Optimization

With good concealment and strong anti-jamming ability,passive tracking has been widely used in military and civilian fields such as intelligence reconnaissance,disaster rescue,unmanned driving and so on.Existing passive tracking methods can be divided into two categories: two-step tracking and direct tracking.The two-step tracking is simple,but its estimation performance is poor in low signal-to-noise ratio.The direct tracking method exhibits good estimation accuracy at the cost of high computational complexity.Many direct tracking algorithms have been proposed in the literature,such as particle filter,Markov chain Monte Carlo,Laplace approximation etc.These methods are all difficult to achieve an fine balance between computational efficiency and estimation accuracy.Therefore,this thesis proposes a new posterior Laplace approximation direct tracking method based on an improved optimization method.Differing from the existing algorithms,it can perform direct tracking quickly and accurately.The main results are listed as follows:(1)Through theoretical derivation and simulation experiments,we show that the performance of direct tracking is better than that of two-step tracking.Firstly,we show that the performance bound of direct tracking is lower than that of two-step tracking under the deterministic state model.Secondly,the performance bounds of two-step tracking and direct tracking are calculated recursively under the nondeterministic state model.Finally,the simulations show that the performance of the two-step tracking algorithm under the two types of state models can not reach that of the direct tracking algorithm;(2)The generalized likelihood Laplace approximation and the posterior Laplace approximation are compared,and the performance advantages of posterior Laplace approximation are analyzed.Firstly,the specific contents of generalized likelihood Laplace approximation and posterior Laplace approximation are depicted,respectively.Secondly,the two algorithms are compared from the three levels of content framework,optimization process and computational complexity.The simulations verify that the posterior Laplace approximation is superior to the generalized likelihood Laplace approximation in terms of tracking accuracy and computational efficiency;(3)A posterior Laplace approximation direct tracking algorithm based on improved optimization is proposed.Firstly,the feature space search technique is given.On this basis,we design an improved optimization algorithm,which uses the feature space search method to help the maximum posteriori optimization algorithm jump out of the neighborhood of the current local maximum to find a better maximum.Finally,the posterior Laplace approximation based on the improved optimization method is proposed.The simulations show that the proposed algorithm greatly improves the tracking accuracy and only slightly increases the computational complexity.