TY - JOUR
T1 - Active target tracking
T2 - A simplified view aligning method for binocular camera model
AU - Li, Xinzhao
AU - Su, Yuanqi
AU - Liu, Yuehu
AU - Zhai, Shaozhuo
AU - Wu, Ying
N1 - Funding Information:
The authors would like to thank all the experts who have contributed to the development of ATT binocular cameras system, especially our colleagues Zhichao Cui, Shiyu Cao, Qiyuan Zhao and Bicheng Wang. Thanks to my wife Jie Ma for her support and dedication. This work is supported by National Natural Science Foundation of China (NSFC) under Grant No. 91520301 and Grant No. 61328303 .
Publisher Copyright:
© 2018
PY - 2018/10
Y1 - 2018/10
N2 - While monitoring large-scale scenes, visual surveillance systems are often confronted with a dilemma between obtaining efficient coverage of the scene and getting sufficient resolution of the targets of interest. To alleviate the contradiction, a solution based on binocular cameras has been proposed. In the solution, an active pan–tilt–zoom (PTZ) camera and a fixed camera operate in a collaborative way by dynamically aligning their fields of view. However, due to the asymmetric camera distribution structure and fast-changing scenario, the accurate alignment is yet challenging. In this paper, a novel view aligning method is presented which introduces three reasonable and mild simplifications to transform the complex view aligning problem into a simple arctangent control function. These simplifications not only relax the restrictions of target depth and the intrinsic parameters of PTZ camera, but also constrain the relative external parameters with a coaxial structure. Meanwhile, a corresponding calibration method is designed to estimate the model parameters off-line. The proposed method was tested in comprehensive cases with relatively close target, wide-range scene and altitude variation. Both the simulation and real scene experiments demonstrate that our method outperforms other state-of-the-art methods on accuracy and effectiveness for active target tracking tasks.
AB - While monitoring large-scale scenes, visual surveillance systems are often confronted with a dilemma between obtaining efficient coverage of the scene and getting sufficient resolution of the targets of interest. To alleviate the contradiction, a solution based on binocular cameras has been proposed. In the solution, an active pan–tilt–zoom (PTZ) camera and a fixed camera operate in a collaborative way by dynamically aligning their fields of view. However, due to the asymmetric camera distribution structure and fast-changing scenario, the accurate alignment is yet challenging. In this paper, a novel view aligning method is presented which introduces three reasonable and mild simplifications to transform the complex view aligning problem into a simple arctangent control function. These simplifications not only relax the restrictions of target depth and the intrinsic parameters of PTZ camera, but also constrain the relative external parameters with a coaxial structure. Meanwhile, a corresponding calibration method is designed to estimate the model parameters off-line. The proposed method was tested in comprehensive cases with relatively close target, wide-range scene and altitude variation. Both the simulation and real scene experiments demonstrate that our method outperforms other state-of-the-art methods on accuracy and effectiveness for active target tracking tasks.
KW - Active target tracking
KW - Camera control
KW - Dual camera system
KW - Video surveillance
KW - View aligning problem
UR - http://www.scopus.com/inward/record.url?scp=85054051586&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85054051586&partnerID=8YFLogxK
U2 - 10.1016/j.cviu.2018.09.005
DO - 10.1016/j.cviu.2018.09.005
M3 - Article
AN - SCOPUS:85054051586
VL - 175
SP - 11
EP - 23
JO - Computer Vision and Image Understanding
JF - Computer Vision and Image Understanding
SN - 1077-3142
ER -