Abstract
Multi-Camera Multi-Target (MCMT) tracking has emerged as a critical capability for applications ranging from autonomous driving to animal behavior monitoring. While recent advances have yielded sophisticated tracking algorithms, the availability of annotated multi-view data remains a significant bottleneck. Existing annotation tools predominantly support single-camera workflows or rely on LiDAR sensors, making cross-view labeling tedious and impractical for camera-only setups.
We present ARGUSTRACK, a multi-camera annotation system that addresses these limitations by enabling annotators to work directly on a bird's-eye-view (BEV) plane. Given calibrated camera parameters, a single ground-plane annotation is automatically projected into 2D bounding boxes across all relevant views, inherently ensuring identity consistency without manual cross-view alignment. To further accelerate the labeling process, ARGUSTRACK incorporates two complementary mechanisms: a Temporal Aware module that propagates annotations from preceding frames to initialize new ones, requiring only minor positional adjustments; and a Multi-camera Semi-annotation module that leverages off-the-shelf 2D detectors combined with foot-point estimation to automatically generate candidate BEV positions for annotator verification.
Video Demos
Normal Annotation
Traditional annotation workflow where annotators label objects on the BEV plane with automatic projection to all camera views.
Temporal Aware Module
The Temporal Aware module propagates annotations from preceding frames, allowing annotators to make minor adjustments instead of labeling from scratch.
Semi-Annotation Module
The Semi-Annotation module automatically generates candidate BEV positions using 2D detectors, shifting the annotator's task from creation to verification.
Comparison with Existing Tools
| Tool | Unified Annotation Space | Low Annotation Time | Cross-view Consistency | Multi-cam Scalability |
|---|---|---|---|---|
| CVAT | ✗ | ✗ | ✗ | ✗ |
| TrackMe | ✗ | ✗ | ✗ | ✗ |
| LabelMe | ✗ | ✗ | ✗ | ✗ |
| ARGUSTRACK (Ours) | ✓ | ✓ | ✓ | ✓ |
Experimental Results
Annotation Speed Comparison
| Tool | Avg. Time / Scene (s) | Speedup |
|---|---|---|
| CVAT | 1872 | 1.0× |
| TrackMe | 1392 | 1.34× |
| ARGUSTRACK (Ours) | 102 | 18.35× |
Ablation Study
| Method | Time (s) |
|---|---|
| ARGUSTRACK (Base) | 150 |
| + Temporal Aware (TA) | 113 |
| + Semi-Annotation (SA) | 102 |
The base ARGUSTRACK pipeline already provides substantial improvement over single-camera workflows by eliminating redundant cross-view labeling. Adding the Temporal Aware module reduces annotation time by propagating annotations from preceding frames. The Semi-Annotation module further reduces time by automatically generating candidate BEV positions from 2D detections, shifting the annotator's task from creation to verification.