图像压缩感知的目标跟踪多特征提取算法  

Multi-Feature Extraction Algorithm Based on Image Compressed Sensing for Target Tracking

作　　者：徐华伟[1] 颜晶晶[1] XU Huawei;YAN Jingjing(Taizhou Vocational & Technical College, Taizhou 318000)

机构地区：[1]台州职业技术学院

出　　处：《计算机与数字工程》2019年第9期2170-2175,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61374148);浙江省教育厅项目“基于大数据分析的国地税异常企业监控平台的研发”(编号:FG2017141)资助

摘　　要：为了提高图像压缩跟踪过程中的目标跟踪性能,论文提出了一种具有随机生成多个特征的目标跟踪算法,通过划分子区域来确定随机生成的多个不同的图像特征进而实现目标跟踪。在每次计算候选位置的最高分类器得分后,结合Bhattacharyya系数来选择最优跟踪结果作为最终目标位置。减少了由图像不良特征确定目标位置的概率,有效地克服了诸如遮挡、形变或类似背景的影响。实验结果表明:相比单个特征提取,两个特征共同提取的中心位置误差平均减少了18.86%,边界框重叠率平均提升了14.14%,成功率平均提升了20.72%,且所提出的跟踪算法的计算速度与图像尺寸有关。In order to improve the performance of target tracking in the process of image compression tracking, this paper proposes a target tracking algorithm with randomly generated multiple features, which determines randomly generated multiple image features and different image features by dividing sub-regions to achieve target tracking. After calculating the highest classifier score for each candidate location, the Bhattacharyya coefficient is used to select the optimal tracking result as the final target location. It reduces the probability of locating the target by bad image features, and effectively overcomes the influence of occlusion, deformation or similar background. The experimental results show that, compared with single feature extraction, the center position error of the two features is reduced by 18.86%, the overlap rate of boundary box is increased by 14.14%, and the success rate is increased by 20.72%. The calculation speed of the proposed tracking algorithm is related to the size of the image.

关 键 词：目标跟踪 特征提取 BHATTACHARYYA系数 图像处理 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]