**4.1. Livestock censors**

To further put **Table 2** in perspective, according to the National Development Agency of South Africa, there were over 13 million units of cattle, 30 million sheep and 6.6 million goats and 1.6 million pigs bred in each province annually between up on to 2003. The figures are even significantly higher in European countries according to Eurostat. These are staggering numbers, hence monitoring and daily head counts of these large number of animals can be challenging. UAVs can thus find application here and be used to perform headcounts of livestock across these large grazing areas [31–33]. Animal counting can be done either by using image recognition [31] or using heat detecting infra-red cameras [34]. For image processing, Convolutional Neural Network (CNN) has emerged in recent times as the most widely used [35]. In large grazing areas, the UAVs can also be used to detect and count the number of animals present. In most of these works, the UAVs fly across the field, and counting the number of animals present. In the work done by [33] however, the authors proposed an approach, wherein the number of goats are counted and tracked using fewer numbers of pictures, sometimes only one. The authors reported 73% count accuracy and 78% tracking accuracy.

In contrast, in their book [34], the authors reviewed numerous methods of performing thermal imaging for monitoring animals in the wild. Among many other factors, the authors argued that thermal imagining is not dependent on time of the day unlike image processing. This therefore provides a unique opportunity to observe animals in their natural habitats without causing disturbances – which can lead to dispersion and possibly double or inaccurate counts.
