Autonomous counting of livestock from remote sensing imagery

Abstract

One of the key issues facing pastoralists across Northern Australia is accurately estimating the number of cattle they have on their property. For smaller producers this has implications in terms of optimising stocking rates to match available resources, thus ensuring sustainability and economic viability. In addition to this, for larger operations, the lack of knowledge about the number of stock has implications for account reporting and ultimately impacts on a number of financial factors including interest rate pricing, costing these operations substantial amounts of money. From a national perspective, the impending emissions trading scheme provides an opportunity for producers to gain benefits from better livestock management, however a lack of information on livestock numbers will certainly limit this. Existing techniques for counting livestock require extensive infrastructure (e.g. camera systems at water points) or devices to be applied to the animal (e.g. RFID tags), all of which are largely impractical solutions for rangeland deployment. In this preliminary study we explored the potential for remotely sensed imagery and image analysis techniques to deliver estimates of livestock populations in a pastoral environment. Airborne imagery was collected using a multispectral system with a spatial resolution of 15cm. A false colour image was developed and used in the subsequent analysis.