Counting Pig using Marker-Controlled Watershed Segmentation

Abstract

Counting the total number of pigs manually on a large-scale pig farm is a crucial and inefficient task. As this process is time-consuming and includes many critical points that can lead to miscalculation. Some of the challenging issues in pig counting include overlapping, partial occlusion, different viewpoint that limits the usage of traditional object detection methods. Image segmentation is used for object detection, which separate foreground and background pixels of the images. In this paper, we used Marker-Controlled Watershed segmentation method for counting pig in an image. Here, different image thresholding techniques such as Otsu threshold, Adaptive threshold and manual threshold is considered. The structural similarity of these thresholding techniques is determined using jaccards coefficient index. Otsu threshold gives the best similarity scores. The average processing time of these thresholding techniques is also determined. Further, the images obtained from Otsu threshold is checked for overlapping objects. In case of image with overlapping objects, the segmentation is done using marker-controlled watershed segmentation algorithm to segregate the overlapping objects and label the objects individually. In case of non overlapping, objects present in the images obtained from Otsu threshold are label directly to count the number of pigs present in the image. Hence, this segmentation process provides an efficient way for counting pigs in an image.