loading page

A Transformer-based Joint Prediction in Hybrid Mechanism with Kalman Filter for 3D-MOT
  • +2
  • Rahmad Sadli ,
  • soheyb Ribouh ,
  • Atika Rivenq ,
  • Abdenour Hadid ,
  • Abdelmalik Taleb Ahmed
Rahmad Sadli
Author Profile
soheyb Ribouh
Univ Rouen Normandie

Corresponding Author:[email protected]

Author Profile
Atika Rivenq
Author Profile
Abdenour Hadid
Author Profile
Abdelmalik Taleb Ahmed
Author Profile

Abstract

This paper presents a novel 3D MOT system exploiting a hybrid mechanism combining Kalman Filters and Transformers for high-performance MOT. We investigate the use of Transformers in 3D Multi-Object Tracking. We propose a new Transformer- based Joint Prediction approach in the Hybrid Mechanism with Kalman Filter for 3D-MOT. To assess the validity of our proposed approach, we compare our experimental results against those of state-of-the-art methods in 3D MOT. Our approach achieves interesting performance and clearly outperforms the PC3T and other methods in terms of tracking precision.