FOOTPRINT RECOGNITION OF RODENTS AND INSECTS Nils Hasler 1 , Reinhard Klette 1 , Bodo Rosenhahn 1 and Warren Agnew 2 1 Centre for Image Technology and Robotics Department of Computer Science, The University of Auckland 2 Connovation Research Ltd. East Tamaki Auckland ABSTRACT In today’s pest control operations large numbers of tracking tunnels are used to estimate the number of rodents present in the target area, providing a basis for planning the required amount of poison. The marks left in the tunnels have to be interpreted by trained experts. This article introduces two methods that make a step towards automating the process of recognizing footprints of rodents and insects. Furthermore two classification methods (Principal Component Analysis, a simple Na¨ ıve Bayes classifier) are studied to distinguish the four examined insect species. Here, a combination of both classifiers proved superior to using just one method. 1. INTRODUCTION AND BACKGROUND We start with a brief overview on the history of rodent erad- ication programs in New Zealand. Then we introduce the approaches to recognize and classify rodent and insect foot- prints in Sections 2 and 3. Section 4 ends with a brief dis- cussion. The four rodent species present in New Zealand are Nor- way rats (Rattus norvegicus), ship rats (Rattus rattus), house mice (Mus musculus or Mus domesticus), and kiore (Rat- tus exulans). With the aid of cats (Felis catus) and stoats (Mustela erminea) they caused the extinction of at least 45 bird species in New Zealand. 1.1. Eradication To counter these developments New Zealand authorities in- troduced eradication programs at the beginning of the last century, first targeting large animals like deer, cattle and goats. Later numbers of smaller animals, namely possums and cats were massively reduced. However, in those first years it was believed that it was impossible to completely eradicate shy, nocturnal rodents from an island. This opinion slowly changed when in the 1970s the second generation of rodenticides became avail- able. In 1982 Ian McFadden was assigned the task to de- velop a method to clear small islands of rodents. With Ruri- ma Rocks in the Bay of Plenty, the new era of rat eradication began [1]. While the first islands were cleared using hand-placed bait stations, slowly the technique evolved to sprinkling the poison across an island and finally to distribution of bait by helicopters. This enabled the Department of Conservation to target larger and larger islands, until in 2001 the largest rat eradication programme to date was carried out on Camp- bell Island, New Zealand. There 200,000 Norway rats were killed on 11,300 hectares using 120 tonnes of poison. In 2003 Campbell Island was officially declared rat-free [2]. Today the eradication techniques developed in New Zealand are exported worldwide for example to the Falkland Islands, Hawaii, and Australia to name just a few. 1.2. Tracking Tunnels Tracking tunnels are basically rectangular polyethylene tun- nels designed to allow the target animal to walk through un- hindered. A tracking card, made of an absorbent white card- board, has an especially developed non-drying ink which has been screened onto a sealed section of the card. On the inked section a lure is placed (e.g., peanut butter for rats and mice). Any animal, or insect, attracted into the tunnel after walking across the ink leaves footprints on the absorbent end section of the card. The oils within the ink are absorbed into the cardboard leaving tracks which can be analysed. They may reveal the genus or species, and possibly the sex of the creature [3]. Tracking tunnels are used to monitor abundance of small mammals. Typically a field study is conducted prior to an eradication operation. Afterwards the study is repeated to check the effectiveness of the implemented procedure. This method was effectively implemented on the Falkland Is- lands [4], on Maui in Hawaii [5] and in various places in New Zealand [6]. Tracking tunnels have also been used to investigate the enormous fluctuations of rodent populations both between seasons and over the years. Fluctuations as high as 90% can be observed between summer and win- ter [7]. This observation leads directly to the question how