Classification of Moving Objects from Real World Image Sequences
Lim, G., Michael D. Alder, Michael D. Alder, C.J.S. deSilva, Yianni Attikiouzel · UWA Profiles and Research Repository (UWA) · 1995
In [1], we presented a syntactic approach to classifying objects in a domestic environment such as human beings, curtains blown by the wind, and external events such as moving tree branches. We introduced Quadratic Neural Networks (QNNs) to model the input data and then extract features of the objects from the model. We called the feature extraction process the UpWrite. With only one level of UpWrite we were able to achieve around 90% recognition rate. In this paper, we introduce another level of UpWrite to overcome some of the problems that the first UpWrite failed to handle. 1 Introduction In [1], we presented a moving object recognition system using a syntactic approach. The syntactic approach to pattern/object recognition decomposes an object into parts, and decomposes each parts into subparts, etc., in a hierarchical manner. The idea is to extract structural information about the objects from the lowest level to the top level, and then perform the classification. At each level, ...