Multi-object tracking (MOT) is a core component of wide-area surveillance systems, where the goal is to accurately follow multiple moving targets over time. A critical challenge in MOT is minimizing identity switches, which occur when the system incorrectly reassigns a tracked object to a different identity. This issue can compromise the integrity of surveillance data, especially in defense or security applications where consistent tracking is essential.
Get the latest gadgets delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
Corvus ISR has published a detailed public tracker benchmark comparing two different models on an identical synthetic scene with perfect ground truth. The scene was designed to test the robustness of each tracker under various stress conditions, with the models evaluated on their ability to maintain correct identities over time. The benchmark highlights how advancements in tracking algorithms can significantly improve performance in complex environments.
The two models under comparison are a baseline “greedy nearest-neighbour” approach (v1) and an advanced “confirmed-track auction” method (v2). The v1 model employs simple association strategies, while v2 uses a sophisticated three-tier auction, velocity-consistency gating, and noise-scaled reservation pricing. These enhancements aim to reduce false reassignments, which are especially problematic in dense scenes with thousands of targets or when facing occlusions and sensor limitations.
Results show that the v2 tracker reduced ID switches per minute by approximately 42% across different scenarios. For instance, in a typical scenario with 150 movers at 2 frames per second, switches dropped from 2,042 to 1,183. The improvements were consistent in denser scenes with 400 movers, where switches decreased from 14,032 to 8,040. The benchmark also tested degraded conditions like lower frame rates and partial occlusions, with reductions of around 18% in switch rates—highlighting the tracker’s resilience.
Importantly, these benchmarks are fully synthetic, generated pixel-by-pixel with perfect ground truth, ensuring the results are measurements rather than marketing hype. Every future tracker submitted must face the same benchmark, fostering transparency in the field. The release aims to push developers toward more reliable tracking solutions, as even the latest models still commit thousands of identity errors per minute under stress conditions.
Engineering-wise, the v2 tracker maintains real-time performance, averaging about 1.2 milliseconds per sensor tick at the highest density—well within typical processing budgets. Anyone can reproduce it live and see these results firsthand. Built with AI automation and independently verified, this synthetic benchmark offers a transparent, accessible way to evaluate multi-object tracking systems without proprietary barriers.
This demonstration underscores a simple truth: in surveillance technology, publicly published failure data fosters genuine progress. Whether in defense, security, or autonomous systems, understanding what works—and what doesn’t—is vital for development. Now, you can test the benchmark yourself and see how your tracker stacks up against the cutting edge.
multi-object tracking surveillance system
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
real-time object tracker for security cameras
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
dense scene object tracking sensor
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
