The science
nesteye combines computer vision, environmental sensing, and edge inference to turn barn activity into practical operating signals.
The goal is not to show operators more data. It is to identify a meaningful change, preserve its context, and make the result measurable.
Piling intelligence
nesteye evaluates flock movement across time. Repeated-frame confirmation helps separate a developing pile from normal clustering, changing light, dust, and the visual complexity of a working barn.
Methods
Each method answers a different operational question while preserving the context needed to interpret the result.
Behavior
A developing cluster must persist across multiple frames before the system treats it as a confirmed event, reducing reactions to ordinary movement.
Growth
Vision-derived body measurements support daily weight and uniformity estimates across more of the flock than a small manual sample.
Context
Environmental readings sit on the same timeline as flock behavior so operators can investigate the conditions surrounding a change.
Diagnostics
Current diagnostic signals are reviewed with flock behavior and barn context so operators can investigate and respond sooner.
Validation
0.998
Piling-detection average precision on held-out test data
This is a test result, not a guarantee of performance in every barn. Layout, installation, lighting, dust, flock behavior, and operating conditions can affect results. Pilot planning establishes the correct baseline for each operation.
Validation boundary
Disease, mortality, and gait diagnostics are current nesteye capabilities. Their signals appear with barn context for operator review, while commercial onboarding establishes barn-specific baselines and validates performance for each operation.
Research foundation
nesteye translates peer-reviewed poultry research into testable barn methods. Published studies guide the architecture; pilot data determines what the product can responsibly claim.
Published work in commercial broiler houses shows that trajectories can be grouped into distinct activity patterns. Our research direction builds on the same principle: persistence, density, and motion change carry more meaning together than any isolated frame.
Read literature review: Campbell et al., 2024Stage 1
Detect
Stage 2
Track
Stage 3
Aggregate
Stage 4
Confirm
Depth-based broiler research combines instance segmentation, physical shape features, and learned features before regression. That informs a careful path toward daily flock weight and uniformity estimates without relying on a few convenient birds.
Read literature review: Li et al., 2023Stage 1
Calibrate
Stage 2
Segment
Stage 3
Measure
Stage 4
Estimate
Automatic piling research compares temporal and convolutional models because dense clustering develops across video, not in a single still. nesteye uses repeated-frame confirmation to distinguish a developing event from ordinary flock movement.
Read literature review: Jensen et al., 2025Stage 1
Density
Stage 2
Duration
Stage 3
Direction
Stage 4
Alert
Evaluate it in context
