Starting from the maybe provocative questions if 'science is mostly driven by ideas or by tools' (Dyson 2012) I will present one of the most difficult computer vision and machine learning scenario, namely quantifying insect behaviour in natural environments: Tiny, low contrast and fast moving objects have to be localised in cluttered and dynamic scenes resulting in visual ambiguities, frequent occlusions and the need to process huge video files with inappropriate foreground-background ratios and sparse visitation events. I will summarize some of the most prominent challenges to detect insects and quantify their in unconstrained settings and will introduce an appearance-agnostic in-field animal detection and tracking algorithm to extract accurate measurements from camera footage. I will close by presenting complementary environmental mapping approaches that yield rich contextual information, and moreover can be combined with behavioural data to achieve the urgently needed lab-level quantifications in wildlife conditions.

Understanding movement of animals is key in ecology. Tracking animals using GPS and accelerometers will not only yield understanding about the whereabouts of wildlife, also about their conservation. I illustrate this by using understanding about predator-prey interactions to develop a system to prevent poaching. I will further explore possible alternatives for GPS to follow wildlife. These examples show that AI is instrumental to developments in ecology.