Abstract: As for target and background changes, this paper proposes a target tracking algorithm with scale and orientation adaptive mean shift tracking with corrected background-weighted histogram algorithms and fast compression tracking algorithms in collaboration based on image classification. According to the difference of the image changes, the algorithm classifies the images into two categories, global changes and target local interest area changes. Global changes caused by lighting, background similar and background blur, it uses BW-SOAMS algorithm for target tracking. Local area of interest changes caused by size, rotate and occlusion, it uses CT algorithm for target tracking. Firstly, images are done preprocessing and classification, and then the appropriate track drift problems are caused by changes in the environment. By experiments, the algorithm has improved significantly in precision and efficiency.

Key words: image classification, robustness, algorithms collaboration, local area of interest, global changes

摘要： 针对目标变化和背景环境的变化，提出了一种基于图像分类的多算法协作的目标跟踪算法，采用融入改进背景加权的尺度方向自适应均值漂移算法与快速压缩算法协作的方式。该算法根据图像变化原因不同将图像分为两类，图像全局变化和目标局部感兴趣区域的变化。对由光照、背景相似度和背景模糊引起的图像全局变化，采用快速压缩算法对目标进行跟踪；对由目标本身尺寸、旋转和遮挡引起的目标局部感兴趣区域变化，采用融入改进背景加权尺度方向自适应均值漂移算法对目标进行跟踪。该算法先对图像序列预处理分类，然后选择适合该对应图像变化特点的算法对目标进行跟踪。经实验验证，该算法较之其他流行目标跟踪算法具有更好的鲁棒性。

关键词: 图像分类, 鲁棒性, 多算法协作, 局部感兴趣区域, 全局变化