Digital image processing methods for the identification of pigs posture during weight estimation

Abstract

Variations in the postures of pigs can contribute to significant amounts of error during automatic weight estimation of pigs using image analysis techniques. This is especially the case when the head of a pig is not maintained on the same straight line as the body of the animal. It was estimated that an additional 2.4% error is associated with sub-optimal posture of animals during weight estimation. To minimize weight estimation error as much as possible, the aim of this study was to identify the most suitable image analysis technique that can automatically select frames from video streams with optimum animal posture for accurate weight estimation. Four different techniques, namely reference line, length ratio, angular width and mid-line methods were investigated using images of pigs between 45 and 90 kg. All images were taken from above the animals at a height of 1680 mm. In this study, images of animals turning their heads more than 30 degrees from the body line either left or right were classified as suboptimal. Similarly, images were also classified as suboptimal if the angular placement of the pigs in the images were more than 30 degrees from the midline. Among the different methods, both the length ratio and angular width method were found to be most likely to identify images with suboptimal head position, while the midline method was the best at identifying pigs with a suboptimal body position. Thus overall the length ratio method appears to be the best to identify suboptimal images before further processing.