基于局部纹理差异性算子的高原鼠兔目标跟踪

Object Tracking of Ochotona curzoniae Based on Local Texture Difference Operator

针对自然生境环境下高原鼠兔目标跟踪中目标与背景颜色相近的问题,提出了一种基于局部纹理差异性算子的高原鼠兔目标跟踪方法。构造了一种新的视觉描述子,称作局部纹理差异性算子LTDC,用来体现目标和背景之间的细微差异性。把该LTDC算子与颜色信息相结合来表征目标模型,并把该目标模型嵌入Meanshift跟踪框架中对高原鼠兔进行跟踪。实验结果表明,所提出的目标表征方法与FLBP目标表征方法相比,具有较强的目标与背景区别能力,在目标和背景颜色相近的场景中,能够较为准确地实现高原鼠兔目标的定位。且所提出的目标表征方法的Meanshift平均迭代次数是FLBP目标表征方法的79.04%,减少了20.96%。跟踪平均总时间是FLBP目标表征方法的82.35%,平均降低了17.65%。同时所提出方法的平均跟踪速度是FLBP目标表征方法的1.22倍。

In order to accurately track Ochotona curzoniae in natural habitat environment, an object tracking method based on Meanshift algorithm was proposed. Considering the object tracking method of kernel Meanshift algorithm based on RGB color histogram usually has the deformation of inaccurate tracking or lose of target in the scenario that the color is similar between the background and the object. In view of the problem that the color between the Ochotona curzoniae and the background is similar in the object tracking process in natural habitat environment, a visual descriptor named as the local texture difference operator （LTDC） was proposed to reflect the subtle differences between the Ochotona curzoniae and background. The LTDC operator was combined with color information to characterize the object model and the object model was embedded into the Meanshift tracking framework for the object tracking of Ochotona curzoniae. The experimental results show that the proposed method for characterizing the object has a strong difference ability of target and background. The object can be accurately positioned in the color similar scenario of the object and the background. Compared with the FLBP algorithm, the average iteration number of proposed method is 79.04% of the average iteration number of the FLBP algorithm, the average tracking total time of proposed method is 82.35% of the average tracking total time of the FLBP algorithm, the average tracking speed of proposed method is 1.22 times of the average tracking speed of the FLBP algorithm