基于HSV颜色特征和贡献度重构的行人跟踪

发布时间：2019-04-24 16:52

【摘要】：在光照、背景变化、遮挡、噪声、快速运动等复杂环境下,准确地实现行人跟踪一直是富有挑战性的任务。针对这些问题,提出基于HSV颜色特征和贡献度重构的行人跟踪算法。在粒子滤波的框架内,从HSV空间提取目标的混合颜色特征生成目标模板集,依据不同区域对跟踪结果的影响对区域进行贡献度分配,并将其引入到一个自适应的正则化模型中,将具有最小重构误差的区域判定为待跟踪目标。为了增强算法的稳健性,跟踪过程中对模板进行实时更新。在OTB 100个序列上进行测试,本文算法得到跟踪结果的平均中心误差和跟踪成功率两项指标分别为0.6624pixel和0.4153,优于同类其他算法。实验结果表明,该算法能够在复杂的视频场景中实现对行人的连续跟踪,且稳健性较好,有利于在实际系统中的实现。
[Abstract]:It is a challenging task to accurately track pedestrians in complex environments such as illumination, background change, occlusion, noise, fast movement and so on. In order to solve these problems, a pedestrian tracking algorithm based on HSV color feature and contribution reconstruction is proposed. In the framework of particle filter, the mixed color feature of the target is extracted from the HSV space to generate the target template set. According to the influence of different regions on the tracking results, the contribution degree of the region is assigned, and it is introduced into an adaptive regularization model. The region with minimum reconstruction error is determined to be the target to be tracked. In order to enhance the robustness of the algorithm, the template is updated in real time during the tracking process. The average center error and the tracking success rate of the algorithm are 0.6624pixel and 0.4153, respectively, which are better than other algorithms of the same kind. The algorithm is tested on 100 sequences of OTB, and the average center error and the tracking success rate of the algorithm are 0.6624pixel and 0.4153 respectively. The experimental results show that the proposed algorithm can continuously track pedestrians in complex video scenes, and the algorithm is robust, which is beneficial to the implementation of the real system.
【作者单位】：首都师范大学信息工程学院;首都师范大学电子系统可靠性技术北京市重点实验室;成像技术北京市高精尖创新中心;首都师范大学高可靠嵌入式系统技术北京市工程技术研究中心;
【基金】：国家自然科学基金(61303104,61203238,11178017,61473015) 北京市自然科学基金(4132014,4162017) 北京市优秀人才资助项目(2016000020124G088)
【分类号】：TP391.41

【相似文献】