为了解决基于颜色直方图的多目标跟踪方法对复杂场景适应能力差,容易丢失目标的问题,文中提出一种将颜色直方图与边缘方向直方图相结合的多目标跟踪方法。该方法首先采用一种分块连通域标记方法进行多目标提取,并获得目标的颜色、边缘特征；然后融合目标颜色与边缘两种特征来描述目标的外观模型；最后对跟踪过程中的目标模板进行更新。实验结果表明,该方法对于目标在尺度、光照、姿态发生变化以及目标发生旋转情况下能够实现目标的稳定跟踪,具有很强的鲁棒性。实验中对3组挑战性的视频序列进行了测试,目标数目选定为2个,目标窗口大小为64 pixels×64 pixels的情况下,本文方法跟踪速度最高可达20 fps,基本上可以满足实时性的跟踪需求。
In order to solve the problem that the color-histogram based multi-target tracking is poorly adaptive to the complex scene and easy to lose the target,a multi-target tracking which combines the color-histogram and edge-direction-histogram is proposed in this paper.Firstly,the algorithm adopts a cell-based connected component labeling algorithm to extract the multiple targets and obtains the color and edge features of the targets.Then we combine the color and edge features to describe the target appearance models.Finally,the target models are updated in the tracking process.The experi-mental results show that the algorithm can complete the stable tracking when the target undergoes the scale variation,illumination changes,pose variation and rotation.And the algorithm has a strong ro-bustness.We perform our algorithm on 3 challenging video sequences in the experiment,when the targets number is chosen to 2 and the target window size is 64 pixels×64 pixels,the maximum speed of our approach can reach 20 fps and it can mainly satisfy the real-time tracking requirement.
参考文献
| [1] | 程帅;孙俊喜;曹永刚;赵立荣.增量深度学习目标跟踪[J].光学精密工程,2015(4):1161-1170. |
| [2] | 李静宇;王延杰.基于子空间的目标跟踪算法研究[J].液晶与显示,2014(4):617-622. |
| [3] | 郭敬明;何昕;魏仲慧.基于在线支持向量机的Mean Shift彩色图像跟踪[J].液晶与显示,2014(1):120-128. |
| [4] | 张雷;王延杰;何舒文.基于相位一致性的实时压缩跟踪方法[J].光子学报,2014(8):124-131. |
| [5] | Comaniciu D.;Ramesh V.;Meer P..Kernel-based object tracking[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,20035(5):564-577. |
| [6] | Zhao Q;Tao H.A motion observable representation using color correlogram and its applications to tracking[J].Computer vision and image understanding: CVIU,20092(2):273-290. |
| [7] | Katja Nummiaro;Esther Koller-Meier;Luc Van Gool.An adaptive color-based particle filter[J].Image and Vision Computing,20031(1):99-110. |
| [8] | He, LF;Chao, YY;Suzuki, K;Wu, KS.Fast connected-component labeling[J].Pattern Recognition: The Journal of the Pattern Recognition Society,20099(9):1977-1987. |
- 下载量()
- 访问量()
- 您的评分：
-
10%
-
20%
-
30%
-
40%
-
50%