.png){kind=icon}
.png){kind=icon}
Return
TMTB: Transformer based multi-task branching multi-object tracking algorithm for wide-view scenes
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-17255-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Combining Unmanned aerial vehicle (UAV) with artificial intelligence can effectively extract information as UAV fly flexibly and have a wide view, but the current processing efficiency of the video information acquired by UAV is low, resulting in insufficient use of resources. In order to enable the tracking algorithm to adapt to wide-angle target scenes and fully utilize the acquired information, we propose a transformer based multi-task branching multi-object tracking algorithm named TMTB. Firstly, a transformer architecture backbone network is designed for extracting target features; then, feature enhancement of multi-stage features through differently focused task branches to suit different objectives in a broad view; finally, the tracking algorithm is optimised according to the motion characteristics of the target in the scene to achieve multi-object tracking in UAV scenes. Experimental results on the VisDrone-MOT2019, and the MOTA is improved by 5.9% compared to JDE, and the operation speed is improved by 75%. The proposed algorithm has good real-time performance and good tracking effect.
Keywords:
Multi-object tracking
Multi-task branches
Wide-view scenes
Unmanned aerial vehicle
Transformer
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W