基于能量的声源目标定位与跟踪

【作者】 于浩；

【导师】 陈雄；

【作者基本信息】 复旦大学 ， 电路与系统， 2008， 硕士

【摘要】 随着无线通信、集成电路、传感器以及微机电系统(MEMS)等技术的飞速发展,无线传感器网络已逐渐成为研究的热点领域。传感器网络主要运用于军事,生态环境监测,基础设施安全,先进制造,物流管理,医疗健康,工业传感,智能交通控制,智能能源等领域,而对进入检测区域的目标进行定位与跟踪是传感器网络的一个重要用途。基于声源的目标定位与跟踪主要有三种方法:TDOA,DOA,能量法。能量法以通信量少,计算方法简单,方便,对整个网络的同步性要求不高等特点,成为本文研究的重点。在基于能量的算法中,能量比例法耗时少,定位速度快,但是定位精度不高,最大似然法虽然定位精度高,但是由于似然函数的非线性,用目前搜索的求解方法,耗时多,定位速度很慢,对目标的定位尤其是需要及时得到目标位置的目标跟踪来说很不利。为了解决定位速度问题,减少定位的耗时,本文在研究了能量模型的最大似然函数的基础上,采用了迭代的方法(主要是高斯-牛顿法)解决能量的似然函数最大值问题,在保证了搜索法定位精度高的前提下,大大提高了定位速度,达到了定位精度和速度的很好的结合。在解决迭代算法的关键问题——局部最小问题上,我们分析了似然函数的特点,提出了在接收能量最大的节点周围选取初始值,来避免迭代走入局部最小的方法,即在接收能量最大的节点周围选取六个初始值进行六次迭代,并将使目标函数最小的那次迭代结果作为目标位置估计值。除此之外,还介绍了如何限制迭代步长,如何确定迭代终止标志,并且将算法推广到三位空间目标的定位与跟踪,证明了在三维空间中新算法同样有效。仿真实验中,迭代算法的精度与搜索法差不多,比能量比例法高很多,而定位时间比搜索法短很多,达到了预期的效果。并且,在目标的跟踪中,由于定位时间短,能够及时得到目标的位置,新算法也得到了比搜索法好的多的结果,尤其在三维中。更多还原

【Abstract】 As the development of wireless communication, IC, sensor and MEMS technology, much more attention is taken to the research on Wireless Sensor Networks(WSN). WSN are widely used in military, circumstance surveillance, infrastructure security, advanced manufacture, goods management, medical treatment, industry sense, intelligent traffic control, intelligent resources and so on. Target localization and tracking is one of the most important applications of WSN.There are three main methods of source localization and tracking, they are TDOA,DOA and Energy. Methods based on energy need less communication, simple computation, and less request on the synchronization of the network, and we mainly research on the methods based on energy in this paper.Of the two methods based on energy, energy ratio(ER) algorithm costs less time, but has a low accuracy. Maximum likelihood estimation has a high accuracy, but as the ML function is non-linear, and by far, search method is the only available resolution to ML function which costs much more time, that is not good for target localization especially for target tracking which need get the target location in time.To improve the localization speed, shorten the cost time, we used iterative algorithm (mainly Gauss-newton) to solve the ML function, after analyzing the character of the function. The new method had as high accuracy as search, but cost much less time, and realized the integration of speed and accuracy.In order to avoid local minimum, based on the analysis of the ML function, we selected initial values around the sensor which received the maximum energy, that is we started six iterative processes from six initial values selected around the sensor. The result which made the function minimum was our estimate value. We also introduced how to control the pace length and confirm when the iterative process stopped. The new algorithm was proved also available in three-dimensional case. It was demonstrated in the experiment that the iterative algorithm was as accurate as the search algorithm, and more accurate than ER, but it cost much less time than search algorithm. That proved our proposal. It performed excellent when tracking a target, especially in three-dimensional case.更多还原