Video analytics automated intake, batch marking, and sex classification at the FAMADESA meat processing plant
At the FAMADESA meat processing plant, the processes of animal intake, batch marking, and sex sorting relied heavily on manual labor.
Project background
At the FAMADESA meat processing plant, the processes of animal intake, batch marking, and sex sorting relied heavily on manual labor. Operators recorded truck arrival times, the number of pigs, vehicle license plates, visually determined the sex of each carcass for correct batch separation, and tracked ear markings to identify the start and end of a batch. This approach led to errors, delays at incoming control, loss of batch data, and reduced overall traceability.
An automated solution was required to determine truck arrival and unloading completion times, automatically count animals, recognize license plates, mark the start and end of a batch based on ear notches, determine sex, and classify organs, with subsequent integration into the production ERP system.
Implementation progress
Specialists from OSMOVIEW LLC developed and implemented a solution for automating intake, marking, and classification at the FAMADESA meat processing plant based on the OSMOCV platform. The platform was deployed on the customer's server infrastructure within a local network.
Video streams from IP cameras installed in key zones were fed into the platform: entrance gates, unloading zone, stunning room, carcass hanging line, and post-mortem inspection zone.
The solution included the following in-house neural services developed by OSMOVIEW LLC:
- License plate recognition (ANPR) — automatic recording of the truck's license plate upon entry for linking to the delivery;
- Animal counting — automatic tracking of pigs during unloading from the vehicle;
- Ear notch classification — recognition of batch start/end markings: no notch, one notch (batch start), two notches (batch end);
- Sex determination based on anatomical features — classification of the notch into two classes: male / female;
- Organ classification — recognition into four classes: male, female, castrated animal, chemically castrated animal.
Additionally implemented:
- automatic linking of the truck number, head count, batch codes, and classification to the production shift;
- development of a custom report aggregating full batch information: vehicle number, arrival and departure times, number of pigs, sex and organ classification — a single document available for viewing and export;
- sending classification results via the MQTT protocol to the customer's production system for further carcass routing;
- configuration of operator alerts for deviations (quantity mismatch, classification anomalies).
Fault-tolerant architecture
A key feature of the project was an architecture with two servers operating in replication mode. The servers mirror each other, ensuring system continuity even if one node fails. This is critical for production, where a stoppage in the accounting and classification system can lead to disruptions in the technological process.
Achieved results
- Reduction in intake time — by 60% due to automatic animal counting, arrival/unloading time recording, and classification without operator involvement.
- Elimination of manual accounting errors — license plate recognition, notch classification, and sex determination are performed with accuracy > 97%, guaranteeing correct batch linking and sorting.
- Full traceability of each batch — the system automatically links the truck's license plate, head count, batch start/end markings, and sex and organ classification.
- Automatic sorting and data transfer — classification results are sent via MQTT to the production system, allowing for the optimization of production lines (separate processing of males, females, castrates, and chemical castrates).
- Custom reporting — a full report is generated for each batch: machine number, arrival and departure times, number of pigs, and classification.
- Increased fault tolerance — two servers in replication mode ensure continuous system operation.
- Statistics collection — by delivery time, loading time, sex distribution, and line performance to optimize logistical and production processes.
- 24/7 operation — the system functions without interruptions, which is particularly important in continuous production environments.