In the realm of **wide-area motion imagery (WAMI)**, achieving accurate **multi-object tracking** remains a significant challenge. These systems must reliably follow hundreds of moving targets across large scenes, where factors like occlusion, jitter, and dense crowds complicate the task. A recent release by CORVUS ISR demonstrates a substantial step forward with their latest tracker models, tested against a rigorous public benchmark.
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The benchmark compares two models: the baseline v1 “greedy nearest-neighbour”, which employs a simple, two-pass greedy association with constant-velocity prediction, and the new v2 “confirmed-track auction”. The latter uses a sophisticated three-tier auction association, velocity-consistency gating, and noise-scaled reservation pricing to improve track stability. These enhancements have yielded a remarkable **42% reduction in identity switches**, which are critical errors where the system incorrectly reassigns a target’s identity.
This benchmark measures **ID switches per minute** across various scenarios, from sparse scenes with 150 movers to dense crowds of 400. Under ideal conditions, v1 recorded 2,042 switches per minute, while v2 cut that number to 1,183 — a clear indicator of increased reliability. Even under challenging conditions, like low frame rate (0.5 fps), occlusions, and sensor degradation, the new tracker consistently outperformed its predecessor.
Why do these **identity errors matter**? In wide-area surveillance, maintaining consistent tracking is vital for intelligence and security applications. The strict ID switch metric used in this benchmark counts every change of identity, including track fragmentations and re-acquisitions, making the improvements even more significant. These results are not marketing hype: they are published measurements on a synthetic scene with perfect ground truth, ensuring transparency.
From an engineering perspective, v2 achieves this with impressive efficiency, averaging around 1.2 milliseconds per sensor tick at a density of 400 targets. Even in the worst cases, it stays within a 5 ms processing window — enabling real-time operation directly in the browser. This accessibility is further emphasized by the fact that anyone can reproduce these results. The live demo allows you to run the benchmark yourself — no sign-up or NDA required.
Built entirely through AI automation, the v2 tracker underwent independent review before release, demonstrating a commitment to transparency and continuous improvement. The synthetic nature of these tests means no real targets or environments are involved, but the measured reductions in identity switches point to promising advancements for future surveillance systems. For those interested in exploring the capabilities firsthand, we invite you to review the public benchmark and reproduce it live to see the results with your own eyes.
wide-area motion imagery tracking system
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
multi-object tracking surveillance camera
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
security surveillance with identity tracking
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
