DiffMOT: A Real-time Diffusion-based Multiple Object Tracker with Non-linear Prediction

Author:

Weiyi Lv, Yuhang Huang, Ning Zhang, Ruei-Sung Lin, Mei Han, Dan Zeng

Keyword:

Computer Science, Computer Vision and Pattern Recognition, Computer Vision and Pattern Recognition (cs.CV)

journal:

date:

2024-03-04 00:00:00

Abstract

In Multiple Object Tracking, objects often exhibit non-linear motion of acceleration and deceleration, with irregular direction changes. Tacking-by-detection (TBD) with Kalman Filter motion prediction works well in pedestrian-dominant scenarios but falls short in complex situations when multiple objects perform non-linear and diverse motion simultaneously. To tackle the complex non-linear motion, we propose a real-time diffusion-based MOT approach named DiffMOT. Specifically, for the motion predictor component, we propose a novel Decoupled Diffusion-based Motion Predictor (D MP). It models the entire distribution of various motion presented by the data as a whole. It also predicts an individual object's motion conditioning on an individual's historical motion information. Furthermore, it optimizes the diffusion process with much less sampling steps. As a MOT tracker, the DiffMOT is real-time at 22.7FPS, and also outperforms the state-of-the-art on DanceTrack and SportsMOT datasets with 63.4 and 76.2 in HOTA metrics, respectively. To the best of our knowledge, DiffMOT is the first to introduce a diffusion probabilistic model into the MOT to tackle non-linear motion prediction.

PDF: DiffMOT: A Real-time Diffusion-based Multiple Object Tracker with Non-linear Prediction.pdf