Tracking Small Birds by Detection Candidate Region Filtering and Detection History-aware Association

TL;DR

The paper addresses the challenge of tracking small, fast-moving birds in panoramic videos where detection and association suffer due to tiny object size. It proposes Adaptive SAHI to selectively sample detection regions around predicted targets and DHSC to perform history-informed, two-stage data association that robustly handles occlusion. Empirical evaluation on the NUBird2022 dataset demonstrates substantial gains in MOTA, IDF1, and FPS, along with dramatic reductions in false positives and identity switches. This approach enables more reliable ecological monitoring of bird behavior in wide-field recordings and can be extended to other small-object tracking tasks and multimodal setups.

Abstract

This paper focuses on tracking birds that appear small in a panoramic video. When the size of the tracked object is small in the image (small object tracking) and move quickly, object detection and association suffers. To address these problems, we propose Adaptive Slicing Aided Hyper Inference (Adaptive SAHI), which reduces the candidate regions to apply detection, and Detection History-aware Similarity Criterion (DHSC), which accurately associates objects in consecutive frames based on the detection history. Experiments on the NUBird2022 dataset verifies the effectiveness of the proposed method by showing improvements in both accuracy and speed.

Figures (6)

• Figure 1: Example of tracking results.
• Figure 2: Object detection with SAHI.
• Figure 3: Process flow of the proposed method.
• Figure 4: Occlusion due to entry and exit.
• Figure 5: Change in number of detection candidate regions.