Detect piling as it forms
Identify dangerous flock density in real time, before compression becomes a mortality event.
NVIDIA Innovation Lab Application
Real-time flock intelligence.Barn-level intervention.
Nesteye combines computer vision, edge AI, and adaptive barn controls to detect risk, estimate flock weight, and intervene before preventable loss occurs.
5Active commercial barn pilots
14Week development program
<30 msTarget inference latency
Edge firstPrivate, resilient deployment
Project objective
Give every barn a live operating picture.
Poultry production still depends on periodic manual observation. Nesteye turns existing visual signals into continuous, actionable intelligence—without disrupting the flock.
Estimate flock weight continuously
Turn overhead imagery into non-contact weight estimates without interrupting normal barn activity.
Surface health risk earlier
Read posture, behavior, movement, and thermal patterns to flag disease and mortality risk.
Intervene inside the barn
Trigger adaptive LED and sound cues that gently disperse piling and help restore healthy movement.
System architecture
From camera frame to action, inside the barn.
The full perception-and-response loop runs locally. Operators get dependable insight even when connectivity is limited, while flock footage stays on site.
01Observe
Barn cameras
Continuous overhead RGB and thermal footage.
02Perceive
Vision models
Segmentation, keypoints, behavior, and posture.
03Infer
Risk engine
Density, weight, health, and mortality signals.
04Act
Barn controls
Adaptive LED and sound intervention.
05Learn
Field feedback
Measured outcomes inform the next model cycle.
Model layer
Custom CNN + YOLOInstance segmentation and piling detection
Keypoints + EfficientNetBody-area features and weight regression
ConvNeXtBehavior, posture, and thermal classification
Data strategy
Train broadly. Validate where the system must work.
A blended data program closes the gap between controlled research and the variability of real commercial barns.
Grounded in poultry research
Data and peer-reviewed findings from the University of Georgia, Texas A&M, and published commercial datasets.
Validated in active barns
Live footage from five commercial pilot barns across Texas and Washington anchors the system in operating conditions.
Expanded through simulation
NVIDIA Omniverse and Isaac Sim vary camera height, angle, lighting, density, and flock behavior at scale.
Accelerated by NVIDIA
Simulation to intervention on one accelerated platform.
NVIDIA technology connects synthetic data generation, model development, video analytics, inference optimization, and edge deployment into one practical path to production.
Why the Innovation Lab
Expert support would accelerate simulation fidelity, model optimization, and field-ready deployment—turning an active pilot system into a scalable commercial platform.
NVIDIA Omniverse
A physically grounded digital barn for repeatable synthetic data generation.
NVIDIA Isaac Sim
Controllable camera, lighting, density, and intervention scenarios.
NVIDIA TAO Toolkit
Transfer learning and efficient fine-tuning for poultry-specific perception.
NVIDIA DeepStream
A real-time, multi-stream video analytics pipeline built for the edge.
NVIDIA TensorRT
Optimized inference with a target latency below 30 milliseconds per frame.
Jetson Orin Nano
Private, resilient in-barn deployment without a dependency on the cloud.
Development plan
Fourteen weeks to a field-validated edge package.
Work moves from simulation and fine-tuning to Jetson optimization and validation in active barns.
1
Weeks 01–02
Environment setup
Configure the simulation environment, ingest source data, and define evaluation baselines.
2
Weeks 03–05
Synthetic data generation
Create diverse barn scenes across camera, lighting, density, and behavior conditions.
3
Weeks 05–09
Model fine-tuning
Train and calibrate segmentation, weight, behavior, posture, and thermal models.
4
Weeks 09–11
Edge optimization
Build the DeepStream pipeline and optimize models with TensorRT for Jetson.
5
Weeks 11–13
Field validation
Measure accuracy, latency, and intervention performance in active commercial barns.
6
Week 14
Finalization
Package the deployment, evaluation report, model assets, and next-stage roadmap.
Program deliverables
A deployable system, measured in real barns.
The program concludes with models, data, validation, and edge deployment assets that can move directly into an expanded pilot.
Download project brief- 01
Working continuous flock-weight estimation model
- 02
Real-time piling detection and intervention system
- 03
Disease and mortality risk detection module
- 04
Reusable synthetic poultry-barn dataset
- 05
Field evaluation and performance report
- 06
Jetson-ready deployment package
Built to extend
One edge system. A growing intelligence layer.
- Longitudinal flock health scoring
- Additional poultry species and barn formats
- Automated welfare and compliance reporting
- Privacy-preserving multi-barn fleet learning