Teat detection algorithm: YOLO vs. Haar-cascade

Abstract

In this study we have developed and experimented with two methods of teat detection based on machine learning approach in image recognition and object detection. Automatic milking systems rely strongly on the vision system for successful milking operation initiation which is the attachment of the teat cups correctly. Teat detection method currently employed in the industry is based on laser assisted edge detection mechanism, making the current systems less advanced than the existing methods in the field of image processing and robotic vision. By experimenting on a basic object detection method based on Haar-like features, viz. Haar cascade classification method and a latest state-of-the-art method based on convolutional neural nets, viz. YOLO-object detection method, we have compared the results of detection on a fake teat model casted from silicon, especially for indoor environments. This study is in extension to the successful real time detection in a cow farm using Haar-cascade based algorithm.