Features values for recognition of swine's head direction from video.

Abstract

Porcine respiratory disease sometime causes the economic damage to swine farmer due to delay of finding and measures. In order to detect infected swine early, farmers observe swine's feeding and drinking movement constantly, therefore automatically observation is useful to reduce the load of farmers. We tried to automatically extract feeding and drinking movement from the video. However, feeding was mis-detected when swine's hip touched feeding area, thus discriminating between head and hip was required. In this paper, we tried to find out features to classify swine images into head or hip, and we evaluated accuracy of classification using the swine area ratio and contour of swine as the features. We performed classification by linear classifier and cross validation with these features. As a result, accuracy of classification was 96.7%, and feature about contour had a different tendency between head and hip.