TL;DR
Corvus ISR has published a reproducible synthetic benchmark reporting that its v2 tracker reduced identity switches by 42.1% with 150 moving objects and 42.7% with 400. The tests use fixed conditions and perfect synthetic ground truth, but the results and review have not been independently verified by an identified third party.
Corvus ISR has published a reproducible synthetic benchmark reporting that its current tracker reduced object identity switches by 42.1% in a 150-mover trial and 42.7% in a denser 400-mover trial. The result matters because identity stability is a central measure of whether a multi-object tracking system can follow the same vehicle or other moving object across successive frames.
In the baseline test, conducted with 150 moving objects at two frames per second, reported identity switches fell from 2,042 to 1,183 per minute. In the dense test with 400 movers, the rate fell from 14,032 to 8,040 switches per minute. Those results compare the v1 tracker, called “greedy nearest-neighbour,” with the v2 tracker, called “confirmed-track auction.”
Corvus ISR says each row uses the same fixed-seed synthetic scene, identified as seed 1337, with a 20-second warm-up and 120 seconds of measurement. According to the published methodology, the sensor model, generated detections and metric definitions remain byte-identical between runs, while only the tracker changes. The company provides archived v1 demonstrations and the current v2 test through a browser-based demo.
The reported gains were smaller under other stresses. Corvus ISR recorded 16.6% fewer switches at 0.5 frames per second, 18.6% fewer with 20% occlusion and 18.1% fewer in a one-frame-per-second test combining jitter with 70% contrast. Detection rates are identical for both tracker versions by construction, according to the benchmark description, because detections are generated by the shared synthetic sensor rather than by either tracker.
Identity Stability Improves Under Load
Identity switches can corrupt an object’s movement history by assigning it a different track label. In a wide-area motion imagery system, fewer switches can produce more stable movement records and reduce errors in downstream analysis. The reported reductions under both baseline and dense traffic suggest that the v2 association method handles competing detections better than the deliberately simple v1 baseline.
The performance figures also address whether the added association logic can run within the demo’s processing limit. At 400-object density, Corvus ISR reports an average of about 1.2 milliseconds per sensor tick and a worst result of roughly five milliseconds against a 10-millisecond budget. If reproduced on comparable hardware, that leaves the browser demonstration operating within its stated real-time allowance.
Multi-Object Tracking by Active Camera
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Fixed Scene Enables Direct Comparison
Corvus ISR is presented as a synthetic WAMI exploitation demonstration. Every person, vehicle, place and pixel is generated, so the benchmark contains no imagery of real subjects. That design gives the test perfect ground-truth identities, allowing each assigned track label to be checked against the known simulated object throughout the run.
The v1 baseline uses two-pass greedy association, constant-velocity prediction and a fixed two-second coasting period. V2 adds track confirmation, a three-tier auction process, velocity-consistency gating, a noise-scaled reservation price and confidence-decayed coasting. The benchmark applies a stricter identity-switch definition than the standard MOTChallenge IDSW measure: fragmentation and reacquisition are counted whenever the track identity assigned to a ground-truth object changes.
“Vendors who show only successes ask for faith; a published failure matrix asks for measurement.”
— Thorsten Meyer AI
DFROBOT HUSKYLENS Smart Vision Sensor for Raspberry Pi, LattePanda or Micro:bit | AI Camera Support Object/Line Tracking, Face/Object/Color/Tag Recognition
HuskyLens is an easy-to-use AI machine vision sensor. It can learn to detect objects, faces, lines, colors and…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Replication Still Pending
The published material does not identify an independent laboratory that reproduced the figures, and it does not provide enough information to compare hardware, browser versions or run-to-run variation across outside systems. Thorsten Meyer AI says the tracker received an independent review before release, but the reviewer, review scope and findings are not disclosed.
The benchmark also does not establish performance on real WAMI footage, where sensor artifacts, imperfect labels and unmodeled behavior may differ from a synthetic scene. Both trackers still produce thousands of switches per minute in demanding rows. The reported percentage gains should be read as improvements against the published v1 floor, not evidence that identity errors have been eliminated.
CAMCAMP Wireless Security Camera System with 10.1" Monitor, Dual Lens PTZ CCTV Camera Security System, 10CH Expandable 4K NVR,Auto Tracking, No Subscription for Outdoor Indoor, 5G & 2.4G WiFi
Dual Lens & Dual View: This dual camera has a fixed bullet camera that stays aimed at the…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Future Trackers Face Same Seed
Corvus ISR says each future tracker version will be added as a new public benchmark row using the same seed and measurement rules. That policy should make later changes directly comparable with v1 and v2, provided the shared sensor and metric code remain unchanged.
The next test of the claim will be outside replication. Users can run the current demo without signup, while researchers would need implementation access and documented runtime conditions to verify the reported reductions and timing results independently.
TP-Link Tapo 1080P Indoor Security Camera for Baby Monitor, Dog Camera w/Motion Detection, 2-Way Audio Siren, Night Vision, Cloud & SD Card Storage, Works w/Alexa & Google Home (Tapo C100)
【Motion Detection & Instant Notification】Get instant push notifications when motion, person or baby crying is detected, there is…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Corvus ISR report?
Corvus ISR reported that its v2 confirmed-track auction tracker reduced identity switches by 42.1% with 150 movers and 42.7% with 400 movers when compared with its v1 greedy baseline.
What is an identity switch?
An identity switch occurs when a tracker assigns a different track identity to the same ground-truth object. Corvus ISR’s metric also counts fragmentation and reacquisition, making it stricter than the cited MOTChallenge definition.
Were real people or vehicles used?
No. The demonstration is described as entirely synthetic, with generated people, vehicles, locations and imagery. This supplies perfect ground truth but does not establish performance on real surveillance footage.
Can the benchmark be reproduced?
The published demo allows visitors to rerun the fixed-seed browser benchmark without signup or an NDA. Independent verification remains limited because the available account does not identify an outside replicator or fully document runtime and review conditions.
Did the new tracker eliminate tracking errors?
No. V2 recorded fewer switches than v1, but it still produced 1,183 switches per minute in the 150-mover baseline and 8,040 per minute in the 400-mover test under the benchmark’s strict counting method.
Source: Thorsten Meyer AI
Source: Thorsten Meyer AI
