SFTrack: A Robust Scale and Motion Adaptive Algorithm for Tracking Small and Fast Moving Objects

TL;DR

The approach involves a new tracking strategy, which initiates the tracking of target objects from low-confidence detections commonly encountered in UAV application scenarios, and proposes revisiting traditional appearance-based matching algorithms to improve the association of low-confidence detections.

Abstract

This paper addresses the problem of multi-object tracking in Unmanned Aerial Vehicle (UAV) footage. It plays a critical role in various UAV applications, including traffic monitoring systems and real-time suspect tracking by the police. However, this task is highly challenging due to the fast motion of UAVs, as well as the small size of target objects in the videos caused by the high-altitude and wide angle views of drones. In this study, we thus introduce a simple yet more effective method compared to previous work to overcome these challenges. Our approach involves a new tracking strategy, which initiates the tracking of target objects from low-confidence detections commonly encountered in UAV application scenarios. Additionally, we propose revisiting traditional appearance-based matching algorithms to improve the association of low-confidence detections. To evaluate the effectiveness of our method, we conducted benchmark evaluations on two UAV-specific datasets (VisDrone2019, UAVDT) and one general object tracking dataset (MOT17). The results demonstrate that our approach surpasses current state-of-the art methodologies, highlighting its robustness and adaptability in diverse tracking environments. Furthermore, we have improved the annotation of the UAVDT dataset by rectifying several errors and addressing omissions found in the original annotations. We will provide this refined version of the dataset to facilitate better benchmarking in the field.

Figures (8)

• Figure 1: Illustration of challenges encountered during multi-object tracking in UAVs. These primarily include irregular patterns of UAV motion, along with low-confidence detections due to small scale objects and occlusions.
• Figure 2: Comparison of tracking results on a low-altitude video. (a) ByteTrack without motion-compensation (MC), fails to maintain tracks. (b) BoTSORT's MC leads to distorted bounding boxes. (c) SFTrack with UAV MC, ensures consistent and accurate tracking.
• Figure 3: This figure compares appearance similarity methodologies for small-scale objects. Re-ID faces challenges in low-resolution and occlusion scenarios, while color histogram and MSE matching offer consistent similarity throughout the frame.
• Figure 4: This image provides a comparative visualization of the Original UAVDT and our Refined UAVDT annotations. Errors in the original annotations, which do not correspond to actual objects, are highlighted in 'red'. The 'yellow' markers represent our additional annotations for visible objects within the Refined UAVDT dataset.
• Figure 5: Experimental comparision with ByteTrack on VisDrone2019 video sequences. Improvements in MOTA and IDF1 scores correlate with higher MRA, indicating the presence of small and fast objects.