Statistical evaluation of dynamical interaction involving bees: bayesian tracking and mutual information

• Autor(a) principal: Oliveira Júnior, Jordão Natal de
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/18/18153/tde-25102021-102129/
• topic: Abelhas de abelhas; Agrochemicals; Agroquímicos; Bayesian inference; Bee swarm Honey bees; Inferência bayesiana; Multi-target tracking; Rastreamento de múltiplos alvos

Tracking objects in video is a cheap method to obtain information about the parts of a system. However, when there are many objects simultaneously in the tracking some problems can happen, such as overlapping and swap of labels, compromising the overall efficiency. Recently new approaches for solving these problems were developed e.g. Convolutional Neural Networks, but the computational cost is still very high. Here, a Bayesian tracking algorithm to supervise objects on video frames is described. The algorithm allows the evaluation of and Probability Distribution Function (PDF) of the objects being tracked by combining the tracking with the Kernel Density Estimation (KDE). The proposed algorithm was evaluated through simulation and comparison with similar approaches, since the conventional databases (as Princeton Tracking Benchmark) lacks similarity with the problem of the one approached in this dissertation. The algorithm is able to track the objects with great precision, thus being able to dynamically evaluate the entropy and energy, by using polar coordinates and assuming a von Mises distribution for the angle variation prediction and a non-informative distribution for the radius prediction. Then, with the information obtained from the algorithm, a resilience analysis was made approaching the effects of two agrochemicals in the honey bees: the insecticide imidacloprid and the fungicide cerconil. Additional information about how they affect honey bees was obtained via Mutual Information on lethal doses, reinforcing the previous results.