Segmentation and Classi cation of Domestic Moving Objects using a Syntactic Approach

In this paper, we propose a method for segmenting and classifying domestic moving objects using a syntactic approach. The problem is to recognise moving objects in front of the camera in a domestic environment such as human beings, curtains blown by the wind and external events such as tree branches, and the aim is to distinguish moving human heads from moving curtains or tree branches. In real-world images, the situation where a human being is moving in the foreground and at the same time the curtains and/or tree branches are moving in the background, often arises. We use quadratic neurons (represented by quadratic forms) as our simple pattern primitives and extract the structural information based on the relationships between the forms. We call this the UpWrite process, and it can be applied any number of times. To achieve our aim of recognising moving heads, we require two levels of the UpWrite process. We hypothesise that the syntactic UpWrite is implemented in wetware by distinct layers of neurons.