This paper presents the results obtained in a real experiment for object recognition in a sequence of images captured by a mobile robot in an indoor environment. The purpose is that the robot learns to identify and locate objects of interest in its environment from samples of different views of the objects taken from video sequences. In this work, objects are simply represented as an unstructured set of spots (image regions) for each frame, which are obtained from the result of an image segmentation algorithm applied on the whole sequence. Each spot is semi-automatically assigned to a class (one of the objects or the background) and different features (color, size and invariant moments) are computed for it. These labeled data are given to a feed-forward neural network which is trained to classify the spots. The results obtained with all the features, several feature subsets and a backward selection method show the feasibility of the approach and point to color as the fundamental feature for discriminative ability.
展开▼
