Aiming at the robustness and efficiency of target representation in tracking process, a new texture description operator is constructed and then a robust tracking algorithm based on it is completed. This algorithm completes texture description in a group of three dimensional vectors by computing the number of pixels in the center similar to that in the neighborhood and making full use of local gradient information of center pixel, then the operator uses the texture information of the true pixels in the target to complete target representation. At the same time, a simple online mechanism to update the target template is adopted. Finally tracking is accomplished by embedding target representation into the framework of particle filter. Experimental results show that proposed algorithm can achieve a robust tracking under the situation of illumination changing, pose changing, occlusion and motion blur, the average tracking success rate is 87.9%, indicating that the operator has good rotation invariance, illumination invariance, and has the strong ability to distinguish complex texture structure.%为解决目标跟踪过程中目标表示鲁棒性和实时性的问题,构建了一种全新的纹理描述算子,并以此完成目标表示进而实现鲁棒的目标跟踪.该算子通过计算邻域像素与中心像素相似的数量,并充分利用中心像素的局部梯度信息,构成一组3维向量完成纹理描述,统计真实目标区域内像素的纹理信息完成目标表示,同时采用一种简单的在线机制更新目标模板,进而嵌入到粒子滤波框架完成目标跟踪.实验结果表明,该方法在光照变化、姿态变化、遮挡、运动模糊等情况下均能完成鲁棒跟踪,平均跟踪成功率为87.9%,表明该算子具有良好的旋转不变性和光照不变性,对复杂纹理结构有较强的区分能力.
展开▼
