Johnson & Johnson

Johnson & Johnson performs animal safety studies in the context of their drug development pipeline. Evaluation of drug‐induced behaviour and (adverse) clinical observations in animals is very important to understand the safety profile of a drug candidate and is performed by well‐trained and experienced lab technicians as a standard approach within the pharmaceutical industry. However, the evaluations by biotechnicians are limited in time and based on subjective interpretation. Hence, adverse events can be missed (e.g. during the night), behaviour in absence of humans is unknown and quantitative analysis of activity and behaviour is lacking. Johnson & Johnson thus turned to the EAVISE research group to develop a pipeline capable of automatically tracking and monitoring dogs.

In the context of this project I managed a small team of researchers to develop the various parts of this pipeline and combine them into a real-time activity tracking application. This project indeed consisted of many different AI models to detect & track the various dogs, recognise their various poses & behaviors, as well as a plethora of clinical observations they might be doing. Combining these different parts together is a non-trivial task, necessitating a uniform pipeline and good communication between the researchers.

Additionally, I developed the dog detector, which lies at the base of the entire pipeline by extracting the coordinates of the dogs in the video frames, as well as the pose classification, enabling the researchers at Johnson & Johnson to gain a better insight in the behavior of the dogs.