UTrack: Multi-Object Tracking with Uncertain Detections

Edgardo Solano-Carrillo,Felix Sattler,Antje Alex,Alexander Klein,Bruno Pereira Costa,Angel Bueno Rodriguez,Jannis Stoppe

2024-08-30

Abstract:The tracking-by-detection paradigm is the mainstream in multi-object tracking, associating tracks to the predictions of an object detector. Although exhibiting uncertainty through a confidence score, these predictions do not capture the entire variability of the inference process. For safety and security critical applications like autonomous driving, surveillance, etc., knowing this predictive uncertainty is essential though. Therefore, we introduce, for the first time, a fast way to obtain the empirical predictive distribution during object detection and incorporate that knowledge in multi-object tracking. Our mechanism can easily be integrated into state-of-the-art trackers, enabling them to fully exploit the uncertainty in the detections. Additionally, novel association methods are introduced that leverage the proposed mechanism. We demonstrate the effectiveness of our contribution on a variety of benchmarks, such as MOT17, MOT20, DanceTrack, and KITTI.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper attempts to address issues primarily focused on uncertainty estimation and utilization in Multi-Object Tracking (MOT). Specifically:

1. **Detection Uncertainty**: Current multi-object tracking methods typically rely on predictions provided by object detectors, which represent uncertainty through confidence scores. However, these confidence scores do not fully capture the variability in the inference process, especially in high-reliability safety and security applications such as autonomous driving and surveillance. Therefore, the paper proposes a method to quickly obtain the prediction distribution during the object detection process, thereby better estimating detection uncertainty.

2. **Utilization of Uncertainty in Tracking**: Existing multi-object tracking methods mainly perform target association through detection results and motion models (such as Kalman filters), but these methods usually do not consider detection uncertainty. The paper proposes new association methods that incorporate detection uncertainty into the tracking process, thereby improving tracking accuracy and robustness.

3. **Real-time Performance**: The proposed method aims to achieve real-time processing and enhance the performance of existing tracking algorithms without significantly increasing computational costs. In particular, the proposed method can be easily integrated into existing state-of-the-art trackers, enabling them to fully utilize the uncertainty information in detection.

In summary, the main goal of the paper is to improve the performance and reliability of multi-object tracking in complex environments by introducing mechanisms for estimating and utilizing detection uncertainty.