基于知识辅助的粒子滤波检测前跟踪算法研究

【作者】 王峰；

【导师】 杨晓波；

【作者基本信息】 电子科技大学 ， 信号与信息处理（专业学位）， 2012， 硕士

【摘要】 隐身技术的出现和超低空突防导致雷达接收回波信号微弱,迫切需要提升雷达的探测能力。检测前跟踪(TBD)技术是通过时间上的观测累积以提高信噪比,能够有效地检测与跟踪微弱目标。粒子滤波(PF)的TBD (PF-TBD)技术适用于非线性非高斯条件下以数值逼近的方法近似估计,成为国内外的研究热点。但是目标的多样性和环境的复杂化限制了PF-TBD算法的探测能力,有效利用先验信息的知识辅助系统能够最大程度地改善TBD技术的探测性能。本文主要是挖掘利用目标与环境的先验信息,结合PF-TBD算法进行研究,其主要内容可以归纳为以下几个方面：1、针对ESIR-TBD算法无法有效地探测强机动性目标的问题,给出了基于多模型的ESIR-TBD算法。该算法通过多种运动模型描述目标运动特性,能够实时选择匹配的运动模型表征其运动模式,有效地检测强机动性目标。同时给出了目标幅度参数的估计方法。2、针对PF-TBD算法在状态空间过大时粒子初始化精度低的问题,利用目标的运动特性,提出了基于竞争机制的粒子初始化方法。该方法通过划分状态子空间,实时更新和选择最优的状态子空间进行初始化,提高了粒子初始化精度。3、针对PF-TBD算法的目标幅度统计模型不匹配问题,利用目标幅度统计分布特性,提出了基于目标幅度信息的PF-TBD算法。该算法有效利用幅度统计特性构造似然比,增强了帧间的幅度相关性,比标准的PF-TBD算法具有更优的检测性能和更高的跟踪精度。4、针对复杂道路场景下的地面移动目标的检测跟踪问题,利用道路交通信息,提出了基于道路交通信息约束的PF-TBD算法。该算法降低目标运动模型的不确定性,约束目标的速度和提高粒子的利用率,提高了微弱目标的探测性能。通过仿真实验的分析,验证了上述方法的有效性。基于知识辅助的粒子滤波TBD算法优于标准的粒子滤波TBD算法,能够更好地探测与跟踪微弱目标。更多还原

【Abstract】 It is an urgent need to improve the detection performance in view of the fact that the radar echoes are much weaker because of stealth technology and over-low altitude penetration. The track-before-detect (TBD) algorithms, accumulating the target energy over time to increase the signal-to-noise ratio (SNR), are therefore highly introduced for the enhancement of the detection performance. Particle filter (PF) based TBD (PF-TBD) algorithm, utilizing the approximable method to estimate the target states under the nonlinear/non-gaossian, is studied widely all over the world. Because the detecting capacity is limited by the growing complexity of the environment and the target variety, the knowledge-based system is used to improve the detection performance.The dissertation focuses on the studies of the PF-TBD algorithm combined with the available priori information. It mainly comprises:1. For solving the problem that it is difficult to effectively detect the high maneuvering target though the ESIR-TBD algorithm, a multiple models based ESIR-TBD algorithm using more models to describe the target movement is proposed. This algorithm can choose the matched model so as to detect the target effectively. Furthermore, the method of estimating the amplitude is provided.2. For solving the problem of lower precision of the particles initialization with the state space being larger for PF-TBD algorithm, a method for particles initialization based on competitive mechanism is proposed by utilizing the moving characteristics for the target. This algorithm can choose the optimal state sub-space and improves precision of the particles initialization though dividing the state sub-space.3. For solving the problem that amplitude statistical model is not matched for PF-TBD algorithm, an amplitude information based PF-TBD algorithm is proposed by incorporating the amplitude information. This algorithm increases the relativity of the neighboring frames for the amplitude by structure new likelihood function with the amplitude information so as to have better detection performance.4. For solving the problem of detecting the ground target under the condition of the complex road network, a novel PF-TBD algorithm is presented by utilizing the road information. This algorithm can improve the detection performance though reducing the dynamic model uncertainty and restraining target velocity.The effectiveness of the methods is validated by simulations. The knowledge-based PF-TBD algorithm outperforms the standard PF-TBD and can detect the weak target effectively.更多还原