A novel probabilistic projection model for multi-camera object tracking
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
Correlation Filter (CF)-based algorithms have achieved remarkable performance in the field of object tracking during past decades. They have great advantages in dense sampling and reduced computational cost due to the usage of circulant matrix. However, present monocular object tracking algorithms can hardly solve fast motion which usually causes tracking failure. In this paper, a novel probabilistic projection model for multi-camera object tracking using two Kinects is proposed. Once the object is found lost using multimodal target detection, the point projection using a probabilistic projection model is processed to get a better tracking position of the targeted object. The projection model works well in the related experiments. Furthermore, when compared with other popular methods, the proposed tracking method grounded on the projection model is demonstrated to be more effective to accommodate the fast motion and achieve better tracking performance to promote robotic autonomy.
|Title of host publication||Towards Autonomous Robotic Systems - 20th Annual Conference, TAROS 2019, Proceedings|
|Editors||Kaspar Althoefer, Jelizaveta Konstantinova, Ketao Zhang|
|Number of pages||11|
|Publication status||Published - Jul 2019|
|Event||20th Annual Conference on Towards Autonomous Robotic Systems - London, United Kingdom|
Duration: 3 Jul 2019 → 5 Jul 2019
|Name||Lecture Notes in Computer Science|
|Conference||20th Annual Conference on Towards Autonomous Robotic Systems|
|Abbreviated title||TAROS 2019|
|Period||3/07/19 → 5/07/19|
Rights statement: The final authenticated version is available online at: http://dx.doi.org/10.1007%2F978-3-030-23807-0_8.
Accepted author manuscript (Post-print), 505 KB, PDF document