基于自适应混合滤波的多目标跟踪算法

Multi-Object Tracking Algorithm Based on Adaptive Mixed Filtering

针对多目标视频跟踪中需要主要解决的目标冲突、合并以及分离等问题,提出了基于自适应混合滤波的多目标跟踪算法。采用混合高斯背景建模法获得前景图,并对图中阴影采用一种简化去除算法,即判断前景像素时,将HSV分量用加权的形式描述,而不必对各个分量依次判断。对前景图提取观测值时,引入了合并处理算法,将分裂的多个矩形检测框进行合并。然后,利用推理的方法将前景观测值与目标关联,用自适应混合滤波算法实现多目标有效跟踪。该算法结合了均值漂移算法运算效率高的和粒子滤波算法能够有效处理遮挡情况的特点。实验表明该算法可以高效地跟踪多目标、准确判断目标的出现和消失,并能够解决多目标冲突、合并和分离等问题。

According to the main problems of multi-object video tracking such as objects collision,merging and splitting,a novel multi-object tracking algorithm based on adaptive mixed filtering is proposed. An adaptive background mixture Gaussian model is adopted to obtain the foreground image,and a simple shadow elimination algorithm is also presented,which describes the HSV components with unified weighted forms,and dose not need judge each component one by one,when it judges the pixels of foreground image. When measured values are extracted from the foreground image,a merging algorithm is introduced,which merges divided detection rectangles into one. Then,the detected foreground measured values are associated with the existing objects based on reasoning methods,and the multiple objects are tracked with adaptive mixed filtering. The algorithm combines the mean shift algorithim which meets the demand of real-time request with the particle filtering one with high reliability when objects are blocked. Simulation experiment proves that the algorithm can track multiple objects efficiently,judge appearance and disappearance of objects accurately,and solve the problems of multi-object blockage,merging and splitting.