Insect vision inspired particle filter for visual tracking

We propose a novel tracking algorithm based on insect vision inspired particle filter. In a cluttered moving background, flying insects demonstrate extraordinary capability in locating and detecting visual objects. Our tracker introduces an Elementary Motion Detector (EMD) which is deduced from the neuronal computational model of the way biological ommateum processing information, and integrates the detector into the particle filter based tracking method. The EMD is utilized as an optimization scheme for proposal distribution in the probabilistic framework of particle filter, where the EMD extracts motion information and responds to dimensional architecture information of motion pattern and particle filter utilizes the motion pattern information to efficiently sample the object's states. An initialization based on motion condition is introduced since the tracked object may disappear and reappear. The idea of the proposed tracking method is simple but effective. Our algorithm is compared with four different tracking methods, and experimental results demonstrate that our method tracks the objects more accurately and reliably in severe tracking environments.