Bioinspired symmetry detection on resource limited embedded platforms

This work is inspired by the vision of flying insects which enables them to detect and locate a set of relevant objects with remarkable effectiveness despite very limited brainpower. The bioinspired approach worked out here focuses on detection of symmetric objects to be performed by resource-limited embedded platforms such as micro air vehicles. Symmetry detection is posed as a pattern matching problem which is solved by an approach based on the use of composite correlation filters. Two variants of the approach are proposed, analysed and tested in which symmetry detection is cast as 1) static and 2) dynamic pattern matching problems. In the static variant, images of objects are input to two dimentional spatial composite correlation filters. In the dynamic variant, a video (resulting from platform motion) is input to a composite correlation filter of which its peak response is used to define symmetry. In both cases, a novel method is used for designing the composite filter templates for symmetry detection. This method significantly reduces the level of detail which needs to be matched to achieve good detection performance. The resulting performance is systematically quantified using the ROC analysis; it is demonstrated that the bioinspired detection approach is better and with a lower computational cost compared to the best state-of-the-art solution hitherto available.