基于背景抑制颜色分布新模型的合成式目标跟踪算法

A Synthetic Target Tracking Algorithm Based on a New Color Distribution Model With Background Suppression

传统的基于直方图分布的目标颜色模型,由于跟踪过程的实时性要求其区间划分不宜过细,因此易导致同一区间有差异的颜色难以区分;此外,还存在易受背景干扰的问题.本文提出一种新的背景抑制目标颜色分布模型,并在此基础上设计了一个合成式的目标跟踪算法.新的颜色分布模型将一阶及二阶统计信息纳入模型,并设计了基于人类视觉特性的权重计算方式,能有效区分同一区间内的差异色且抑制背景颜色在模型中的比重;算法基于该颜色模型构建目标的产生式模型,并引入结合方向梯度直方图(Histogram of oriented gradient,HOG)特征的相关滤波器对目标形状进行判别式建模,同时将两个模型相互融合;针对融合参数不易设计的难点,分析并建立了一套定性原则,用于判定模型各自的可信度并指导模型更新;最终利用粒子群算法的搜索机制对候选目标的位置、尺度进行搜索,其中适应值函数设计为两个跟踪器的融合结果.实验结果表明,本文算法在绝大多数情况下准确率较对比算法更优且能满足实时性要求.

The traditional histogram distribution based target color model easily fails to discern colors within a color interval that cannot be too small due to the real-time tracking requirement.Moreover,the model is prone to background interference.In this paper,a new target color distribution model with background suppression is proposed and synthetic target tracking algorithm based on the new model is presented.The new model takes the first-and second-order statistical information into account and makes use of human visual characteristics for weight computation.This approach allows to distinguish different colors within the same color interval and to suppress the proportion of background colors in the target model.The proposed algorithm builds up a target generative model on the basis of the new color model and figures out a target shape discriminative model using the correlation filter in terms of histogram of oriented gradient(HOG)features.These two models are then fused for target tracking.A set of qualitative principles for setting fusion parameters is given for evaluating the individual confidence of both models and for model updating.Finally,a particle swarm optimization algorithm is applied to identify the locations and scales of candidate targets using a fitness function determined by the tracking result of the fusion model.The experimental results show that the proposed algorithm in most cases outperforms the other algorithms in terms of accuracy while meeting the requirement of real time tracking.