Development and application of computer vision systems for use in livestock production

Abstract

This paper examines the feasibility of applying computer vision systems to improve health, welfare and efficiency in livestock production. Very little directly relevant literature was revealed when reviewing the subject, so it is examined from first principles. After briefly describing the value of computer vision as a sensor with powerful observational and interpretative ability, the different steps in vision system development are identified and explored. Where possible this examination is related to computer vision work on livestock as well as other biological objects which by their typically varied nature offer meaningful paradigms for the livestockrelated work. The analysis suggests that most operations in livestock production tend to be at the complex end of the spectrum of vision-related problems currently being tackled in agriculture. Hence, only applications which have a significant production or welfare effect will be viable. Another vital element necessary for success in this application is a simultaneous understanding by the system designers of a diverse set of mechanisms (the production process, the interaction between process and sensor, vision algorithm building, and software and hardware systems). This calls for a multi-disciplinary, interactive approach to develop optimal solutions.