Predictive State Estimation of Invasive Predators using Low Resolution Thermal Cameras
Ben McEwen, Richard Green, Stefanie Gutschmidt, Grant Ryan · 2021
Thermal cameras are used to monitor invasive pest populations and inform elimination efforts. These cameras are limited by their resolution, meaning that feature-based classification is often not sufficient. Predictive state estimation of invasive predators is useful for visual classification and the analysis of movement patterns in occluded and noisy environments. These movement patterns aid in the classification of species. Multiple State estimation techniques, such as the Kalman Filter, Unscented Kalman Filter and Particle Filter, were tested on a thermal recording dataset. The state estimation techniques were compared using the thermal dataset and it was found that they were able to improve tracking performance in noisy and occluded environments with the Unscented Kalman Filter achieving the best results. It was found that these methods all suffer from similar limitations due to the animal changing state while occluded. A potential solution to these limitations is proposed.