Multisensor Integration for Scene Classifiction: An Experiment in Human Form Detection

This paper presents a system for classiication of scenes using a multisensor integration framework. Indoor scenes are imaged using a visual and an infrared sensor and the images processed in three stages to perform classiication of sensed objects into two classes: human and background. Finally, information from individual classiiers is integrated in order to obtain an improved classiication performance. Details of feature extraction and classiication using neural network combining are presented. Segmentation of the imaged scene is performed using existing techniques such as texture analysis and histogram modeling. Classiica-tion results on real-world data are presented. The system represents a rst step in the development of improved , robust classiiers based on the concepts of neu-ral networks and multisensor integration.