A Biologically Inspired Optical Flow System: an Application for Motion Segmentation and Object Identification

Optical flow is possibly the most used method for motion segmentation. However its application is often restricted to off-line processing as it requires extensive computational resources and time. In this work, we explore an optical flow method derived from research on the vision system of diptereous insects. The proposed method, Biological Optical Flow (BioOF) was implemented using a series of filters, and therefore is much faster than any existing machine-coded optical flow algorithm. Like other optical flow methods, the output of the BioOF has two components: horizontal and vertical optical flows -- both of them are combined in order to get a better final result in terms of motion segmentation. The result is a framework that can extract an excellent contour of the moving objects segmented out from the images. Finally the object contour is projected onto a Fourier feature space, leading to a representation of the object that is rotational and translational invariant. Over the Fourier feature space, various classification algorithms are investigated for object recognition.