On the Efficacy of Appearance and Geometric Features for Outdoor Scene Classification

This paper presents a comparison of the performance of different combinations of features for the composite classification of outdoor urban workspaces. In line with our previous work the performance evaluation is based on the application of support vector machines. Feature combinations of varying information richness are compared for all workspace classes considered using receiver operating characteristics. We argue that, across all classes, a basic combination of a single 3D geometric feature and basic colour features is a competitive alternative to more elaborate feature sets