New WAMI Tracker Cuts Identity Switches by 42% in Synthetic Tests

AIThis post was created with the assistance of artificial intelligence (AI).

Corvus ISR tracker benchmark matrix (seed 1337) — The published matrix — every row reproducible. Source: corvusisr.com/benchmark

In the realm of wide-area motion imagery (WAMI), accurately tracking multiple moving objects remains a significant challenge. Corvus ISR, a leading provider in this space, has recently published a detailed public tracker benchmark comparing two sophisticated models on a synthetic scene with perfect ground truth. This benchmark is designed to evaluate how well different algorithms maintain consistent identities across frames, a critical aspect for surveillance and intelligence operations.

The study pits the baseline v1 model, based on a two-pass greedy association with constant-velocity prediction, against the newer v2 model that employs an auction-based confirmation process. The latter integrates three-tier auction association, velocity-consistency gating, and confidence-decayed coasting, significantly improving the tracker’s ability to retain object identities even in dense or occluded situations.

Results speak volumes: under standard conditions with 150 movers at 2 frames per second, the v2 model reduced ID switches per minute from 2,042 to just 1,183—a reduction of over 42%. Similar improvements were observed with increased density, dropping from 14,032 to 8,040 switches (−42.7%). These metrics are especially strict, counting every change of the assigned identity, including re-acquisitions and fragmentations, making the improvements all the more notable.

Understanding why these numbers matter is key: in wide-area surveillance, maintaining correct identities over time is essential. Mistakes like ID switches can lead to misinterpretations or missed targets. Corvus ISR’s decision to publish these failure metrics openly emphasizes the importance of measuring progress transparently, especially since the scenes are synthetic with perfect ground truth—these are real performance indicators, not marketing claims.

Corvus ISR live demo — The live demo — press “Run benchmark” to reproduce the numbers. Source: corvusisr.com/demo

Importantly, the v2 tracker runs in real-time within a browser environment, processing roughly 1.2 milliseconds per sensor tick at the highest density tested (400 objects), with a worst case around 5ms against a 10ms budget. Anyone interested can easily reproduce it live—there’s no signup or NDA required, just click and run the benchmark. This democratizes access to cutting-edge multi-object tracking technology and invites developers to experiment firsthand.

Built by an AI executor and independently reviewed before release, the v2 model exemplifies how automation and rigorous testing are shaping the future of surveillance tech. With every pixel synthetic, the benchmarks provide a clear, measurable pathway for future improvements, emphasizing that even the best models still generate thousands of identity errors per minute under stress. This transparency encourages ongoing innovation and precise development in multi-object tracking systems.