Estudo comparativo de técnicas de detecção do espelho nasal de bovinos

Abstract

Bovine identification is very important in many circumstances, such as controlling disease outbreaks and managing cattle production. One of the means by which identification can be done is by analyzing an animal’s muzzle print features, as extracted out of a photo containing the animal. Under this context, one of the subproblems is detecting the muzzle region in a photo that also has many other objects. This work compares two techniques that can be used for object detection in this context: Cascade Classifier using MB-LBP Features and YOLO (You Only Look Once) CNN based Classifier. In general, the YOLO detector had better results in all scenarios and managed a region of interest false positive detection rate of 0.00%.