Automatic validation of animal vaccination on video streams for Merial
Merial, a world leader in animal health, is seeking to automate part of its processes, particularly in the field of vaccination.
In this context, Astrakhan intervened in a project aimed at automatically detecting that animals are properly vaccinated after injection by a robot, in order to limit the number of manual checks.
Implementation of an automatic detection method allowing to give a probability that the vaccine was correctly injected for each of the animals shown in a video feed, each animal having been injected with a vaccine (with success if blue pellet visible on the animal’s body) :
• Automatic animal detection in the videos
• Extraction of different visual characteristics on each animal (feature engineering) using, among other things, colour filters
• Implementation of a training dataset annotated by Merial experts (each animal labelled as “vaccinated” or “not vaccinated”)
• Training of different Machine Learning algorithms from training data and selection of the best algorithm
• Automation of prediction of the trained algorithm on video streams
• Notebook program to easily create a training dataset by modifying color filters
• Script creating a serialized file of a trained Machine Learning model
• Main program that can be run on a continuous video stream and gives vaccination probabilities for each identified animal, by choosing a trained Machine Learning model beforehand
Our added value
• Machine Learning expertise
• Features engineering expertise
• Computer Vision knowledge