融合颜色特征和灰度特征的时空上下文跟踪

Spatio-temporal context tracking combined with color attributes and gray feature

时空上下文视频目标跟踪算法仅使用单一灰度特征,跟踪精度低,在面内外旋转情况下容易跟踪失败。针对该问题,提出一种改进的特征融合算法。考虑到颜色特征的尺度不变性和光照平衡性,引入能更好表示目标外观的11种基础色表示特征。提取目标及其局部上下文的11维颜色特征并进行自适应降维,融合降维后的颜色特征与灰度特征构建目标外观模型,在线学习并更新时空模型,计算置信图并寻找其最大值作为目标位置输出。仿真结果表明,改进算法提高了原算法在光照变化、遮挡等情况下的跟踪精度,能够处理面内外旋转的跟踪难题。

In traditional spatio-temporal context(STC)tracking,only a single gray-scale feature is used.It has low tracking precision and is easy to fail when there are plane rotations.Aiming at this problem,an improved STC algorithm was proposed in which gray feature and color attributes were fused.Taking into account the scale invariance and the light balance of the color feature,11 kinds of basic color were introduced to express the appearance of a target.The 11 dimensional color features of target and the local context were extracted and the former features were reduced to 2 dimensions using an adaptive method.The target appearance model was constructed by combining the reduced dimensional color feature and the gray feature.The spatio-temporal context model was learned and updated.The confidence map was calculated and the maximal value was found out to be the location of target.Experimental results show that the improved algorithm can deal with the tracking problem of plane rotation and improve the accuracy of the STC algorithm when illumination change and occlusion occur.