Recipient Organization
INNOVATIVE AUTOMATION TECHNOLOGIES, LLC
1805 SW 131ST ST
NEWBERRY,FL 32669
Performing Department
(N/A)
Non Technical Summary
The CoopAEye project vision is development of a low-cost camera system which provides full commercial poultry house visual coverage with edge AI processing capabilities. The project vision is validated through a proof-of-concept which utilizes a representative hardware setup, real and synthesized data, and focuses on activity observations.The CoopAEye distinguishes itself from other systems by providing complete poultry house imaging and enabling detection/tracking of every bird. This is achieved through a distributed mesh of inexpensive camera/processor nodes which perform edge image processing and wireless synchronization/communication. During Phase I, IATech develops a Proof of Concept of the CoopAEye product, a low-cost camera system specifically for enabling complete coverage of commercial poultry houses. This provides growers with total poultry house imagery and frequent animal observations such as activity.Potential benefits of CoopAEye:Allows complete imaging of commercial poultry house livestockallows for animal observation and condition monitoring of poultry without having to enter the facility thereby reducing the risk of contamination to the flock or the farmerlife cycle data collection for production analysisproactive data driven managementenables subtle detection of changes to flock dynamics for early problem tracingcondensed animal analytics for archival compared to recorded live video.
Animal Health Component
34%
Research Effort Categories
Basic
33%
Applied
34%
Developmental
33%
Goals / Objectives
The goals for this project include:Demonstrationof processing and sensor hardware/software andassociated communication, camera synchronization, and power distribution functionalityImplemention ofrealistic poultry house simulationDemonstration ofartificial intelligencealgorithms for automated animal detection and activity measurementIdentification ofstandards and best practices for operational and cyber securityObjective 1: Embedded Hardware and Infrastructure Risk Reduction ResearchFor this objective our team performs development of custom embedded hardware, software, and infrastructure to enable the CoopAEye system. Our team uses Commercial Off The Shelf (COTS) components to create the baseline product and conducts a cost analysis to determine the system cost at various levels of production. The resulting analysis assesses the economies of scale for the CoopAEye system and answers the following questions:What is the production rate required to achieve a system cost under $2500 for an average size poultry house installation?Are there components/subassemblies that can be procured to reduce cost instead of re-designing and producing in-house?How low can the price be driven down and what volume is required?What is the price comparison between domestic and international production?Objective 2: AI Training Risk Reduction ResearchThis objective demonstrates the feasibility of using poultry house digital twin simulations for training and testing of the CoopAEye system. The outcome of this objective is a high quality simulation of a poultry house structure and animals which provides AI algorithm training and input for embedded processor testing. This research will answer the following questions:How is the simulation output optimized to match AI algorithm input requirements?How are the activity levels of the autonomous agents (simulated poultry) simulated and databased for later comparison?Is there specialized hardware/software that can improve performance of the simulation for researchers?Objective 3: Full Scale Poultry House Coverage Risk Reduction ResearchThis objective demonstrates full scale poultry house coverage by constructing a proof-of-concept CoopAEye system. The outcome of this objective is a sub-scale installation that can capture, and process simulated and real poultry observations and answer the following questions:How is the system designed to simplify installations (self-calibrating)?How is the system self-healing when camera nodes need to be replaced?How are the camera nodes synchronized so that imagery is taken at the same time?How are overlapping images and activity observations combined to form a single cohesive data set?How can existing poultry house framework be used to support camera node installation?Objective 4: Cyber Security Risk Reduction ResearchThis objective establishes key information of potential system exploits and provides a framework for future system implementation. Industry experts have identified system integrity and security as one of the key risks of this technology. As part of this objective, this research will answer the following questions:What are the potential attack vectors?What hardware, software, and industry practices can be implemented to reduce cyber intrusion risk?What are best practices from other industries which can be used to improve cyber security of the CoopAEye hardware and software?
Project Methods
Task 1: Cost AnalysisFor this task, a cost analysis is performed to estimate the cost of the CoopAEye system for an average size poultry house of 20k ft2 (40ft x 500ft x 10ft). Many new houses in use today are larger than this, however, we are using this size as a baseline for this analysis. The CoopAEye system can be scaled to support larger or smaller houses. Using the current estimate of 200 computing modules and cameras to provide full coverage, and communication/power infrastructure, the procurement and production costs are estimated for low and high volume production. Price breaks for critical components are documented and are used for tiered production calculation. Innovative use of commercial off the shelf sub-assemblies is also investigated for potential value-added features and cost savings.Task 2: Poultry House Digital Twin SimulationBeing able to detect and track the location of every bird in a poultry production facility has the potential to gain knowledge of individual bird conditions and flock conditions. Previous research has determined that proper spatial distribution of poultry is an indication of a healthy flock [Guo et al. 2020]. As part of the simulation, our team incorporates elements from the gaming industry to create a realistic multi-agent simulation. The environment and agents of the simulation are realistically modeled and serve to provide input data for the hardware in the loop testing. The poultry are modeled after straight run Ross 708 poultry stock, at 4 weeks of age. Variation is size and weight of male and female are incorporated into the model. Dynamics and kinematics of the skeletal structure are guided by established gait score assessment criteria. The goal is for the simulated poultry to be scored by a trained assessor at the same level as the input.The housing model is developed based on a 20k ft2 (40ft x 500ft x 10ft) structure. Components like waterers and feeders are modeled in CAD (Solidworks) and exported to compatible file formats.This simulated data takes the place of hand sampling video data and inspection by veterinarians in order to extract training data sets. The simulated data is inspected and validated by veterinarians and subject matter experts to ensure the simulated birds and actions look and "behave" like real birds.Realistic rendering is conducted using Unreal Engine (Unreal Engine 5) and is performed on a Dell Precision Tower 3620 Workstation. The physics engine and multi-agent automation software is developed using GCC to allow for porting across platforms (Windows/Linux).Task 3: Hardware in the Loop TestingA camera and computing cluster of 100 imaging/computing nodes is constructed for the hardware in the loop testing. An 8020 framework is constructed inside of the IATech manufacturing facility to position the camera/processor node array and support wiring. The poultry house digital twin simulation generates frames for each node based on the poultry states, and camera position. The YOLO object detection algorithm is used on each computing node to detect poultry in the image frames and translate to spatial position and velocity. A Dell XPS laptop aggregates post processed data from each node and databases full house stitched imagery and tracking/activity data.In order to facilitate camera calibration and self healing, fiducial targets (April tags) are placed strategically throughout the real and simulated environment. This allows for the camera nodes to calculate their position without having to be precisely placed or oriented.Task 4: Cyber Security AnalysisFor this task, our team examines the interface and communication infrastructure of the baseline system. NIST 800-171 is used as guidance for the analysis. Additional cyber security best practices are incorporated to establish a risk mitigation plan for further development. In addition, procedures and hardware/software tools are proposed to improve operational and cyber intrusion prevention.