107 © Institution of Engineers Australia, 2011 * Reviewed and revised version of a paper originally presented at the 2009 Society for Engineering in Agriculture (SEAg) National Conference, Brisbane, Queensland, 13-16 September 2009. † Corresponding author A/Prof Thomas Banhazi can be contacted at thomas.banhazi@usq.edu.au. Improved image analysis based system to reliably predict the live weight of pigs on farm: Preliminary results * TM Banhazi † and M Tscharke National Centre for Engineering in Agriculture, University of Southern Queensland, Toowoomba, Queensland WM Ferdous, C Saunders and S-H Lee Department of Mechanical Engineering, University of South Australia, Mawson Lakes, South Australia ABSTRACT: A computer vision system was developed to automatically measure the live weight of pigs without human intervention. The system was trialled on both research and commercial farms to demonstrate the ability of the system to cope with different conditions and non-uniform lighting conditions. Early results demonstrate that the system can achieve suficient practical accuracy. The results of the initial trials demonstrated that weight of the pigs can be predicted with an average error of 1.18 kg. Precision, reliability and repeatability are likely to be increased in future through improved weight prediction models, increased image resolution and algorithm enhancement. 1 INTRODUCTION It is desirable to monitor the weight of domesticated animals regularly as it gives an indication of the animals’ rate of growth. An animal’s growth rate is important as it ultimately determines the proitability of the farming enterprise (Schofield, 1990). The weight of the livestock can also be associated with size, shape and condition of the animal (Brandl & Jorgensen, 1996). Therefore, it is desirable to monitor the weight of domesticated animals regularly (Schoield et al, 2002). Traditionally, weighing of livestock is performed manually on-farm using mechanical or electronic scales. This practice is labour intensive, time consuming and potentially dangerous (Kollis et al, 2007). Furthermore, the procedures associated with manual weighing are stressful for the animals (Kollis et al, 2007). As a result of these dificulties several research groups around the world are exploring alternative weighing methods with the aim of overcoming the previously mentioned problems on commercial farms. One non-contact animal weighing method that research groups are currently exploring is video image analysis (VIA). This weighing method predicts a live animals weight based on body measurements extracted from a video frame (Doeschl-Wilson et al, 2005; Schoield et al, 1999; Whittemore & Schoield, 2000; Wouters et al, 1990). The body measurements can then be modelled to predict the weight, size and body composition of the animal to reasonable accuracy (Wang et al, 2006; Schoield et al, 1999; Doeschl-Wilson et al, 2005). Two of the earliest applications of computer vision in the area of live weight analysis were conducted in the UK and Denmark. Research performed at the Silsoe Institute (UK) in the late eighties demonstrated that IA can be used to predict the weight of pigs within 5% accuracy based on a pigs body area (Schoield, 1990; Schoield et al, 1999). Subsequently the VIA-based weighing system developed at the institute was used as part of a prototype closed-loop, model-based control system of pig growth. The core of the system was a mechanistic growth model which monitored environmental and animal variables as inputs for control, such as room ammonia levels and the pigs’ growth. A trial demonstrated that the system has the potential to control the growth rate and condition of a group of pigs (Parsons et al, 2007). A similar model based control system is currently being developed for the Australian pig industry (Banhazi et al, 2007a; 2007b; Banhazi & Black, 2009). Another VIA-related experiment was conducted in the mid-1990s in Denmark. In this experiment 416 crossbred pigs were weighed and captured on video. Samples were taken ive times during their life time between approximately 25 and 100 kg live-weight. Australian Journal of Multi-disciplinary Engineering, Vol 8 No 2