基于阈值修正的自适应新生目标状态提取

Adaptive newborn target state extraction based on threshold correction

传统的多目标跟踪算法采取在量测附近进行粒子采样来近似表示新生目标,虽然这种方法有一定的可操作性,但在量测似然较大时,该量测更有可能来自存活目标。对此,采用自适应新生目标粒子PHD滤波方法,这种方法将存活目标和新生目标作为两个不同的部分分别进行滤波。同时,由于聚类K均值算法进行PHD状态提取时对初始粒子的选取有很强的依赖性,并且容易受到噪声的干扰,影响状态提取的结果,为了解决这个问题,在自适应新生目标强度PHD滤波的基础之上,设置阈值筛选权值较大的粒子进行归一化处理,这种方法大大简化了目标状态提取的过程,进一步提高了滤波的精度。

Traditional multiple target tracking algorithm usually carries out particle sampling near the measurement to estimate the newborn target approximately,although this method has some maneuverability,it has large chance comes from survival target when the measurement likelihood is high. The adaptive newborn target particle PHD filter is adopted in this paper,which treats survival targets and newborn targets as two different filtering parts. Because the K-means clustering algorithm has very strong dependence to the initial particle selection and susceptible to noise,the extraction results is affected. In order to solve this problem,based on the adaptive newborn target PHD filter,this paper sets the threshold to select particles with larger weight. This method could greatly simplify the process of target state extraction,further improve the filtering accuracy.