Liu, K.-C.K.-C.LiuShen, Y.-T.Y.-T.ShenLIANG-GEE CHEN2020-06-112020-06-112018https://scholars.lib.ntu.edu.tw/handle/123456789/499324https://www.scopus.com/inward/record.uri?eid=2-s2.0-85048808883&doi=10.1109%2fICCE.2018.8326132&partnerID=40&md5=4c82046eb1132e6f12607a5a38fd058cIn this paper, a simple yet effective method for online and real-time multi-object tracking (MOT) in 360-degree equi-rectangular panoramic videos is proposed. Based on the current state-of-the-art tracking-by-detection paradigm, several improvements have been made to overcome the challenges of full field-of-view (FOV) of Spherical Panoramic Camera (SPC). In addition, two datasets are presented for evaluation. It is shown that the proposed method outperforms the baseline by 28.6% and 27.8% in terms of average Multiple Object Tracking Accuracy (MOTA) on each dataset. © 2018 IEEE.convolution neural network; equirectangular panorama; multi-object tracking; online and realtime tracking; spherical panoramic cameraCameras; Spheres; Convolution neural network; Equirectangular panoramas; Multi-object tracking; Panoramic cameras; Real time tracking; Object trackingconference paper10.1109/ICCE.2018.83261322-s2.0-85048808883