A compressive tracking based on time-space Kalman fusion model
The compressive tracking (CT)method is a simple yet efficient algorithm which compresses the high-dimensional features into a low-dimensional space while preserving most of the salient information. This paper proposes a compressive time-space Kalman fusion tracking algorithm to extend the CT method...
Saved in:
Published in | Science China. Information sciences Vol. 59; no. 1; pp. 117 - 131 |
---|---|
Main Authors | , , , , |
Format | Journal Article |
Language | English |
Published |
Beijing
Science China Press
01.01.2016
Springer Nature B.V |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The compressive tracking (CT)method is a simple yet efficient algorithm which compresses the high-dimensional features into a low-dimensional space while preserving most of the salient information. This paper proposes a compressive time-space Kalman fusion tracking algorithm to extend the CT method to the case of multi-sensor fusion tracking. Existing fusion trackers deal with multi-sensor features individually and without time-space adaptability. Besides, significant information cumulated in the updating process has not been fully exploited, which calls for a necessity for temporal information extraction. Unlike previous algorithms, the proposed fusion model is completed in both space and time domains. Also, extended Kalman filter is introduced to formulate an updating method for fusion coefficient optimization. The accuracy and robustness of the proposed fusion tracking algorithm are demonstrated by several experimental results. |
---|---|
Bibliography: | 11-5847/TP fusion tracking, compressive tracking, time-space Kalman fusion model, extended Kalman filter,visual tracking The compressive tracking (CT)method is a simple yet efficient algorithm which compresses the high-dimensional features into a low-dimensional space while preserving most of the salient information. This paper proposes a compressive time-space Kalman fusion tracking algorithm to extend the CT method to the case of multi-sensor fusion tracking. Existing fusion trackers deal with multi-sensor features individually and without time-space adaptability. Besides, significant information cumulated in the updating process has not been fully exploited, which calls for a necessity for temporal information extraction. Unlike previous algorithms, the proposed fusion model is completed in both space and time domains. Also, extended Kalman filter is introduced to formulate an updating method for fusion coefficient optimization. The accuracy and robustness of the proposed fusion tracking algorithm are demonstrated by several experimental results. |
ISSN: | 1674-733X 1869-1919 |
DOI: | 10.1007/s11432-015-5356-0 |