.png){kind=icon}
.png){kind=icon}
Return
Real-Time Multi-object Pedestrian Tracking Method Based on Dynamic Matching Weights
.png){kind=icon}
PRE
AI
DOI：10.1007/978-981-95-4821-7_25.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The Detection and Tracking (TBD) framework demonstrates excellent tracking accuracy due to its robust target localization capabilities. However, in complex environments, dynamic backgrounds, lighting changes, and occlusions lead to target loss, significantly increasing the number of identity switches (IDS) in multi-object tracking (MOT) across frames. To address this issue, this paper proposes a method for dynamically adjusting matching weights by using a time function to dynamically adjust the weights of short-term IoU matching and long-term appearance matching, thereby improving the recovery capability after target loss. Experimental results show that the proposed method achieves further improvements in IDF1 and IDS metrics on the MOT16 dataset, with IDF1 reaching 70.8, an increase of 4.4 percentage points compared to static weights, and IDS reduced by 38%. This clearly demonstrates that, in complex dynamic environments, the proposed dynamic weight adjustment algorithm effectively balances the importance of IoU matching and appearance information matching.
Keywords:
detection tracking framework
tmulti-object tracking
tdynamic matching weight
tfeature extraction
Journal
A
IF:
0
Papers:
26
Citations:
0