激光与光电子学进展, 2017, 54 (9): 091006, 网络出版: 2017-09-06
基于隐马尔可夫模型和分块特征匹配的目标跟踪算法
Object Tracking Algorithm Based on Hidden Markov Model and Block Feature Matching
图像处理 目标跟踪 主成分分析 尺度变化 Camshift算法 隐马尔可夫模型 image processing object tracking principal component analysis scale variation Camshift algorithm hidden Markov model
摘要
为解决运动目标跟踪过程中由于遮挡、光照变化、尺度变化等因素导致的目标易丢失以及传统Camshift跟踪算法中跟踪窗口易发散等问题, 提出一种融合优化的隐马尔可夫模型(HMM)和分块特征匹配的运动目标跟踪算法。首先, 利用主成分分析(PCA)结合特征位置对目标仿射尺度不变特征变换(ASIFT)特征进行降维生成PCA-ASIFT特征, 保留目标关键信息; 其次, 采用粒子滤波最优特征位置优化目标PCA-ASIFT特征的HMM参数; 最后, 通过HSV直方图模型建立目标分块, 赋予不同目标分块相应权重并结合分块特征匹配以改善Camshift算法实现运动目标跟踪。实验结果表明, 在自然场景下, 本文算法能够取得较好的运动目标跟踪效果, 对遮挡、尺度变化等具有较好的稳健性。
Abstract
In the process of moving object tracking, in order to solve the problems that the object is easy to loss because of the occlusion, illumination fluctuation, scale variation and other factors, and the tracking window of the traditional Camshift algorithm is easy to diverge, a moving object tracking algorithm is proposed based on the fusion of optimized hidden Markov model (HMM) and the block feature matching. Firstly, the principal component analysis (PCA) combined with the feature position is used to reduce the dimension of the affine scale invariant feature transformation (ASIFT) features to generate PCA-ASIFT features which can retain the key information of the object. Then, the of the PCA-ASIFT features can be optimized by using the optimal feature positions of the particle filter. Finally, the object blocks are established by HSV histogram model and the different weights are assigned to different blocks and the integration block features matching, which can improve the Camshift algorithm to accomplish the moving object tracking. The experimental results show that the proposed algorithm can achieve better tracking effect of moving object in natural scenes, and it has bette robustness to occlusion, scale variation and so on.
陆兵, 顾苏杭. 基于隐马尔可夫模型和分块特征匹配的目标跟踪算法[J]. 激光与光电子学进展, 2017, 54(9): 091006. Lu Bing, Gu Suhang. Object Tracking Algorithm Based on Hidden Markov Model and Block Feature Matching[J]. Laser & Optoelectronics Progress, 2017, 54(9): 091006.