'Circular Health for Industry': Object Detection and Tracking for animals' social interactions

This thesis is based on a project, called ’Circular Health for Industry’ developed by myself together with the Business Intelligence team of Alten Italia S.p.A.. The main goal of this project was, given different cameras in a stable for cows, to detect and track cows to understand social relations, habits and correlation of these factors with their milking quantity and quality. In the first chapter we are going to face a little introduction about artificial neural network, in particular convolutional neural network. They will be the main architectures above which we will build the mathematical models. In the second chapter we will discuss two object detectors we implemented and trained in the project: ’Single Shot Detector’ and ’EfficientDet’. The third chapter will be about the tracking problem, in particular we will see how ’Simple Online Real-Time Tracking’ works. To conclude we developed a final chapter with future developments of the project, in particular regarding the projection problem from image pixels to real word coordinates.