Image vision based AI to predict foaling and send alarm to horse owners
Horse owners always want to participate when a horse is having a foal but it't very difficult to predict when foaling is about to happen. This leads to many sleepless nights.
Our Image Vision model learns the behavior of the mare, sends the behavior values to our algorithms that has been trained to detect signs of foaling. The end product is a phone call to the horse owner saying that the horse is probably about to have a foal within the next hours.
Problem description
In the mare, gestational length is highly variable (320-360 days), as are biochemical, anatomical and clinical signs of imminent foaling. Further, the majority of foalings take place during the night. This makes parturition hard to predict, and manual surveillance labor-heavy. Should a problem occur, quick intervention is needed to save foal and mare. Reliable systems that will alert the person when foaling is approaching are therefore sought after. Today, many different systems are used, with varying success.
Objective
Develop an AI-system that based on image vision and algorithms can detect a behavior that indicates that the horse is about to have a foal soon. The system should include everything from hardware, image vision model, app, alarm system and supporting infrastructure.
Hardware
Rasberry Pi 3b computer together with a camera module is build in a “camera unit” specially designed for stables, including an IR-lamp for vision during night time.
Image Vision and analysis
We are using semantic segmentation model to detect horses in the box then apply our custom classification model to detect its behavior. In order to aggregate and improve AI results, relying on our statistical knowledge from historical data we are applying several rules to initial results.
An end-to-end solution with an AI-camera that is installed in the stable, learns the mare's behavior and calls the owner before foaling is about to happen. Results from 2021's foalings were a successful alarm in 94,4% of the foalings that were monitored.
It is a launched product and the customers are paying for the foaling alarm service and also consumes the service of having the camera and the app as a combination of sleep and activity watch but for horses.
